Digital Video Processing

Digital Video Processing
Digital Video Processing
Second Edition
This page intentionally left blank
Digital Video Processing
Second Edition
A. Murat Tekalp
New York • Boston • Indianapolis • San Francisco
Toronto • Montreal • London • Munich • Paris • Madrid
Capetown • Sydney • Tokyo • Singapore • Mexico City
Many of the designations used by manufacturers and sellers to distinguish their products are claimed as trademarks.
Where those designations appear in this book, and the publisher was aware of a trademark claim, the designations
have been printed with initial capital letters or in all capitals.
The author and publisher have taken care in the preparation of this book, but make no expressed or implied
warranty of any kind and assume no responsibility for errors or omissions. No liability is assumed for incidental or
consequential damages in connection with or arising out of the use of the information or programs contained herein.
For information about buying this title in bulk quantities, or for special sales opportunities (which may include
electronic versions; custom cover designs; and content particular to your business, training goals, marketing focus, or
branding interests), please contact our corporate sales department at [email protected] or (800) 382-3419.
For government sales inquiries, please contact [email protected] For questions about sales outside the United States, please contact [email protected] Visit us on the Web:
Library of Congress Cataloging-in-Publication Data
Tekalp, A. Murat.
Digital video processing / A. Murat Tekalp.—Second edition.
pages cm
Includes bibliographical references and index.
ISBN 978-0-13-399100-0 (hardcover : alk. paper)—ISBN 0-13-399100-8 (hardcover : alk. paper)
1. Digital video—Textbooks. I. Title.
TK6680.5.T45 2015
Copyright © 2015 Pearson Education, Inc.
All rights reserved. Printed in the United States of America. This publication is protected by copyright, and
permission must be obtained from the publisher prior to any prohibited reproduction, storage in a retrieval system,
or transmission in any form or by any means, electronic, mechanical, photocopying, recording, or likewise. To
obtain permission to use material from this work, please submit a written request to Pearson Education, Inc.,
Permissions Department, 200 Old Tappan Road, Old Tappan, New Jersey 07675, or you may fax your request to
(201) 236-3290.
ISBN-13: 978-0-13-399100-0
ISBN-10: 0-13-399100-8
Text printed in the United States on recycled paper at Courier in Westford, Massachusetts.
First printing, June 2015
To Sevim and Kaya Tekalp, my mom and dad,
To Özge, my beloved wife, and
To Engin Deniz, my son, and Derya Cansu, my daughter
This page intentionally left blank
Preface xvii
About the Author xxv
1 Multi-Dimensional Signals and Systems 1
1.1 Multi-Dimensional Signals 2
1.1.1 Finite-Extent Signals and Periodic Signals 2
1.1.2 Symmetric Signals 5
1.1.3 Special Multi-Dimensional Signals 5
1.2 Multi-Dimensional Transforms 8
1.2.1 Fourier Transform of Continuous Signals 8
1.2.2 Fourier Transform of Discrete Signals 12
1.2.3 Discrete Fourier Transform (DFT) 14
1.2.4 Discrete Cosine Transform (DCT) 18
1.3 Multi-Dimensional Systems 20
1.3.1 Impulse Response and 2D Convolution 20
1.3.2 Frequency Response 23
1.3.3 FIR Filters and Symmetry 25
1.3.4 IIR Filters and Partial Difference Equations 27
1.4 Multi-Dimensional Sampling Theory 30
1.4.1 Sampling on a Lattice 30
1.4.2 Spectrum of Signals Sampled on a Lattice 34
1.4.3 Nyquist Criterion for Sampling on a Lattice 36
1.4.4 Reconstruction from Samples on a Lattice 41
1.5 Sampling Structure Conversion 42
References 47
Exercises 48
Problem Set 1 48
MATLAB Exercises 50
2 Digital Images and Video 53
2.1 Human Visual System and Color 54
2.1.1 Color Vision and Models 54
2.1.2 Contrast Sensitivity 57
2.1.3 Spatio-Temporal Frequency Response 59
2.1.4 Stereo/Depth Perception 62
2.2 Analog Video 63
2.2.1 Progressive vs. Interlaced Scanning 64
2.2.2 Analog-Video Signal Formats 65
2.2.3 Analog-to-Digital Conversion 66
2.3 Digital Video 67
2.3.1 Spatial Resolution and Frame Rate 67
2.3.2 Color, Dynamic Range, and Bit-Depth 69
2.3.3 Color Image Processing 71
2.3.4 Digital-Video Standards 74
2.4 3D Video 79
2.4.1 3D-Display Technologies 79
2.4.2 Stereoscopic Video 82
2.4.3 Multi-View Video 83
2.5 Digital-Video Applications 85
2.5.1 Digital TV 85
2.5.2 Digital Cinema 89
2.5.3 Video Streaming over the Internet 92
2.5.4 Computer Vision and Scene/Activity Understanding 95
2.6 Image and Video Quality 96
2.6.1 Visual Artifacts 96
2.6.2 Subjective Quality Assessment 97
2.6.3 Objective Quality Assessment 98
References 100
3 Image Filtering 105
3.1 Image Smoothing 106
3.1.1 Linear Shift-Invariant Low-Pass Filtering 106
3.1.2 Bi-Lateral Filtering 109
3.2 Image Re-Sampling and Multi-Resolution Representations 110
3.2.1 Image Decimation 111
3.2.2 Interpolation 113
3.2.3 Multi-Resolution Pyramid Representations 120
3.2.4 Wavelet Representations 121
3.3 Image-Gradient Estimation, Edge and Feature Detection 127
3.3.1 Estimation of the Image Gradient 128
3.3.2 Estimation of the Laplacian 132
3.3.3 Canny Edge Detection 134
3.3.4 Harris Corner Detection 135
3.4 Image Enhancement 137
3.4.1 Pixel-Based Contrast Enhancement 137
3.4.2 Spatial Filtering for Tone Mapping and Image Sharpening 142
3.5 Image Denoising 147
3.5.1 Image and Noise Models 148
3.5.2 Linear Space-Invariant Filters in the DFT Domain 150
3.5.3 Local Adaptive Filtering 153
3.5.4 Nonlinear Filtering: Order-Statistics, Wavelet Shrinkage,
and Bi-Lateral Filtering 158
3.5.5 Non-Local Filtering: NL-Means and BM3D 162
3.6 Image Restoration 164
3.6.1 Blur Models 165
3.6.2 Restoration of Images Degraded by Linear Space-Invariant
Blurs 169
3.6.3 Blind Restoration – Blur Identification 175
3.6.4 Restoration of Images Degraded by Space-Varying Blurs 177
3.6.5 Image In-Painting 180
References 181
Exercises 186
Problem Set 3 186
MATLAB Exercises 189
MATLAB Resources 193
4 Motion Estimation 195
4.1 Image Formation 196
4.1.1 Camera Models 196
4.1.2 Photometric Effects of 3D Motion 201
4.2 Motion Models 202
4.2.1 Projected Motion vs. Apparent Motion 203
4.2.2 Projected 3D Rigid-Motion Models 207
4.2.3 2D Apparent-Motion Models 210
4.3 2D Apparent-Motion Estimation 214
4.3.1 Sparse Correspondence, Optical-Flow Estimation, and
Image-Registration Problems 214
4.3.2 Optical-Flow Equation and Normal Flow 217
4.3.3 Displaced-Frame Difference 219
4.3.4 Motion Estimation is Ill-Posed: Occlusion and Aperture
Problems 220
4.3.5 Hierarchical Motion Estimation 223
4.3.6 Performance Measures for Motion Estimation 224
4.4 Differential Methods 225
4.4.1 Lukas–Kanade Method 225
4.4.2 Horn–Schunk Motion Estimation 230
4.5 Matching Methods 233
4.5.1 Basic Block-Matching 234
4.5.2 Variable-Size Block-Matching 238
4.5.3 Hierarchical Block-Matching 240
4.5.4 Generalized Block-Matching – Local Deformable Motion 241
4.5.5 Homography Estimation from Feature Correspondences 243
4.6 Nonlinear Optimization Methods 245
4.6.1 Pel-Recursive Motion Estimation 245
4.6.2 Bayesian Motion Estimation 247
4.7 Transform-Domain Methods 249
4.7.1 Phase-Correlation Method 249
4.7.2 Space-Frequency Spectral Methods 251
4.8 3D Motion and Structure Estimation 251
4.8.1 Camera Calibration 252
4.8.2 Affine Reconstruction 253
4.8.3 Projective Reconstruction 255
4.8.4 Euclidean Reconstruction 260
4.8.5 Planar-Parallax and Relative Affine Structure
Reconstruction 261
4.8.6 Dense Structure from Stereo 263
References 263
Exercises 268
Problem Set 4 268
MATLAB Exercises 270
MATLAB Resources 272
5 Video Segmentation and Tracking 273
5.1 Image Segmentation 275
5.1.1 Thresholding 275
5.1.2 Clustering 277
5.1.3 Bayesian Methods 281
5.1.4 Graph-Based Methods 285
5.1.5 Active-Contour Models 287
5.2 Change Detection 289
5.2.1 Shot-Boundary Detection 289
5.2.2 Background Subtraction 291
5.3 Motion Segmentation 298
5.3.1 Dominant-Motion Segmentation 299
5.3.2 Multiple-Motion Segmentation 302
5.3.3 Region-Based Motion Segmentation: Fusion of Color and
Motion 311
5.3.4 Simultaneous Motion Estimation and Segmentation 313
5.4 Motion Tracking 317
5.4.1 Graph-Based Spatio-Temporal Segmentation and Tracking 319
5.4.2 Kanade–Lucas–Tomasi Tracking 319
5.4.3 Mean-Shift Tracking 321
5.4.4 Particle-Filter Tracking 323
5.4.5 Active-Contour Tracking 325
5.4.6 2D-Mesh Tracking 327
5.5 Image and Video Matting 328
5.6 Performance Evaluation 330
References 331
MATLAB Exercises 338
Internet Resources 339
6 Video Filtering 341
6.1 Theory of Spatio-Temporal Filtering 342
6.1.1 Frequency Spectrum of Video 342
6.1.2 Motion-Adaptive Filtering 345
6.1.3 Motion-Compensated Filtering 345
6.2 Video-Format Conversion 349
6.2.1 Down-Conversion 351
6.2.2 De-Interlacing 355
6.2.3 Frame-Rate Conversion 361
6.3 Multi-Frame Noise Filtering 367
6.3.1 Motion-Adaptive Noise Filtering 367
6.3.2 Motion-Compensated Noise Filtering 369
6.4 Multi-Frame Restoration 374
6.4.1 Multi-Frame Modeling 375
6.4.2 Multi-Frame Wiener Restoration 375
6.5 Multi-Frame Super-Resolution 377
6.5.1 What Is Super-Resolution? 378
6.5.2 Modeling Low-Resolution Sampling 381
6.5.3 Super-Resolution in the Frequency Domain 386
6.5.4 Multi-Frame Spatial-Domain Methods 389
References 394
Exercises 399
Problem Set 6 399
MATLAB Exercises 400
7 Image Compression 401
7.1 Basics of Image Compression 402
7.1.1 Information Theoretic Concepts 402
7.1.2 Elements of Image-Compression Systems 405
7.1.3 Quantization 406
7.1.4 Symbol Coding 409
7.1.5 Huffman Coding 410
7.1.6 Arithmetic Coding 414
7.2 Lossless Image Compression 417
7.2.1 Bit-Plane Coding 418
7.2.2 Run-Length Coding and ITU G3/G4 Standards 419
7.2.3 Adaptive Arithmetic Coding and JBIG 423
7.2.4 Early Lossless Predictive Coding 424
7.2.5 JPEG-LS Standard 426
7.2.6 Lempel–Ziv Coding 430
7.3 Discrete-Cosine Transform Coding and JPEG 431
7.3.1 Discrete-Cosine Transform 432
7.3.2 ISO JPEG Standard 434
7.3.3 Encoder Control and Compression Artifacts 442
7.4 Wavelet-Transform Coding and JPEG2000 443
7.4.1 Wavelet Transform and Choice of Filters 443
7.4.2 ISO JPEG2000 Standard 448
References 454
Exercises 456
Internet Resources 459
8 Video Compression 461
8.1 Video-Compression Approaches 462
8.1.1 Intra-Frame Compression, Motion JPEG 2000, and
Digital Cinema 462
8.1.2 3D-Transform Coding 463
8.1.3 Motion-Compensated Transform Coding 466
8.2 Early Video-Compression Standards 467
8.2.1 ISO and ITU Standards 467
8.2.2 MPEG-1 Standard 468
8.2.3 MPEG-2 Standard 476
8.3 MPEG-4 AVC/ITU-T H.264 Standard 483
8.3.1 Input-Video Formats and Data Structure 484
8.3.2 Intra-Prediction 485
8.3.3 Motion Compensation 486
8.3.4 Transform 488
8.3.5 Other Tools and Improvements 489
8.4 High-Efficiency Video-Coding (HEVC) Standard 491
8.4.1 Video-Input Format and Data Structure 491
8.4.2 Coding-Tree Units 492
8.4.3 Tools for Parallel Encoding/Decoding 493
8.4.4 Other Tools and Improvements 495
8.5 Scalable-Video Compression 497
8.5.1 Temporal Scalability 498
8.5.2 Spatial Scalability 499
8.5.3 Quality (SNR) Scalability 500
8.5.4 Hybrid Scalability 502
8.6 Stereo and Multi-View Video Compression 502
8.6.1 Frame-Compatible Stereo-Video Compression 503
8.6.2 Stereo and Multi-View Video-Coding Extensions of
the H.264/AVC Standard 504
8.6.3 Multi-View Video Plus Depth Compression 507
References 512
Exercises 514
Internet Resources 515
A Vector-Matrix Operations in Image and Video Processing 517
A.1 Two-Dimensional Convolution 517
A.2 Two-Dimensional Discrete-Fourier Transform 520
A.2.1 Diagonalization of Block-Circulant Matrices 521
A.3 Three-Dimensional Rotation – Rotation Matrix 521
A.3.1 Euler Angles 522
A.3.2 Rotation About an Arbitrary Axis 523
A.3.3 Quaternion Representation 524
References 525
Exercises 525
B Ill-Posed Problems in Image and Video Processing 527
B.1 Image Representations 527
B.1.1 Deterministic Framework – Function/Vector Spaces 527
B.1.2 Bayesian Framework – Random Fields 528
B.2 Overview of Image Models 528
B.3 Basics of Sparse-Image Modeling 530
B.4 Well-Posed Formulations of Ill-Posed Problems 531
B.4.1 Constrained-Optimization Problem 531
B.4.2 Bayesian-Estimation Problem 532
References 532
C Markov and Gibbs Random Fields 533
C.1 Equivalence of Markov Random Fields and Gibbs Random Fields 533
C.1.1 Markov Random Fields 534
C.1.2 Gibbs Random Fields 535
C.1.3 Equivalence of MRF and GRF 536
C.2 Gibbs Distribution as an a priori PDF Model 537
C.3Computation of Local Conditional Probabilities from a Gibbs
Distribution 538
References 539
D Optimization Methods 541
D.1 Gradient-Based Optimization 542
D.1.1 Steepest-Descent Method 542
D.1.2 Newton–Raphson Method 543
D.2 Simulated Annealing 544
D.2.1 Metropolis Algorithm 545
D.2.2 Gibbs Sampler 546
D.3 Greedy Methods 547
D.3.1 Iterated Conditional Modes 547
D.3.2 Mean-Field Annealing 548
D.3.3 Highest Confidence First 548
References 549
E Model Fitting 551
E.1 Least-Squares Fitting 551
E.2 Least-Squares Solution of Homogeneous Linear Equations 552
E.2.1 Alternate Derivation 553
E.3 Total Least-Squares Fitting 554
E.4 Random-Sample Consensus (RANSAC) 556
References 556
Index 557
This page intentionally left blank
The first edition of this book (1995) was the first comprehensive textbook on digital
video processing. However, digital video technologies and video processing algorithms were not mature enough then. Digital TV standards were just being written,
digital cinema was not even in consideration, and digital video cameras and DVD
were just entering the market. Hence, the first edition contained some now-outdated
methods/algorithms and technologies compared with the state of the art today, and
obviously missed important developments in the last 20 years. The first edition was
organized into 25 smaller chapters on what were then conceived to be important
topics in video processing, each intended to be covered in one or two lectures during
a one-semester course. Some methods covered in the first edition—e.g., pel-recursive
motion estimation, vector quantization, fractal compression, and model-based
coding—no longer reflect the state of the art. Some technologies covered in the first
edition, such as analog video/TV and 128K videophone, are now obsolete.
In the 20 years since the first edition, digital video has become ubiquitous in our
daily lives in the digital age. Video processing algorithms have become more mature
with significant new advances made by signal processing and computer vision communities, and the most popular and successful techniques and algorithms for different tasks have become clearer. Hence, it is now the right time for an updated edition
of the book. This book aims to fill the need for a comprehensive, rigorous, and
tutorial-style textbook for digital image and video processing that covers the most
recent state of the art in a well-balanced manner.
This second edition significantly improves the organization of the material
and presentation style and updates the technical content with the most up-to-date
techniques, successful algorithms, and most recent knowledge in the field. It is
organized into eight comprehensive chapters, where each covers a major subject,
including multi-dimensional signal processing, image/video basics, image filtering,
motion estimation, video segmentation, video filtering, image compression, and
video compression, with an emphasis on the most successful techniques in each
subject area. Therefore, this is not an incremental revision—it is almost a complete
The book is intended as a quantitative textbook for advanced undergraduateand graduate-level classes on digital image and video processing. It assumes familiarity with calculus, linear algebra, probability, and some basic digital signal processing
concepts. Readers with a computer science background who may not be familiar
with the fundamental signal processing concepts can skip Chapter 1 and still follow
the remaining chapters reasonably well. Although the presentation is rigorous, it is in
a tutorial style starting from fundamentals. Hence, it can also be used as a reference
book or for self-study by researchers and engineers in the industry or in academia.
This book enables the reader to
understand theoretical foundations of image and video processing methods,
learn the most popular and successful algorithms to solve common image and
video processing problems,
reinforce their understanding by solving problems at the end of each chapter, and
practice methods by doing the MATLAB projects at the end of each chapter.
Digital video processing refers to manipulation of the digital video bitstream.
All digital video applications require compression. In addition, they may benefit
from filtering for format conversion, enhancement, restoration, and super-resolution in order to obtain better-quality images or to extract specific information, and
some may require additional processing for motion estimation, video segmentation, and 3D scene analysis. What makes digital video processing different from still
image processing is that video contains a significant amount of temporal correlation
(redundancy) between the frames. One may attempt to process video as a sequence
of still images, where each frame is processed independently. However, multi-frame
processing techniques using inter-frame correlations enable us to develop more effective algorithms, such as motion-compensated filtering and prediction. In addition,
some tasks, such as motion estimation or the analysis of a time-varying scene, obviously cannot be performed on the basis of a single image.
It is the goal of this book to provide the reader with the mathematical basis of
image (single-frame) and video (multi-frame) processing methods. In particular, this
book answers the following fundamental questions:
How do we separate images (signal) from noise?
Is there a relationship between interpolation, restoration, and super-resolution?
How do we estimate 2D and 3D motion for different applications?
How do we segment images and video into regions of interest?
How do we track objects in video?
Is video filtering a better-posed problem than image filtering?
What makes super-resolution possible?
Can we obtain a high-quality still image from a video clip?
What makes image and video compression possible?
How do we compress images and video?
What are the most recent international standards for image/video compression?
What are the most recent standards for 3D video representation and compression?
Most image and video processing problems are ill-posed (underdetermined and/
or sensitive to noise) and their solutions rely on some sort of image and video models. Approaches to image modeling for solution of ill-posed problems are discussed in
Appendix B. In particular, image models can be classified as those based on
local smoothness,
sparseness in a transform domain, and
non-local self-similarity.
Most image processing algorithms employ one or more of these models. Video
models use, in addition to the above,
global or block translation motion,
parametric motion,
motion (spatial) smoothness,
motion uniformity in time (temporal continuity or smoothness), and
planar support in 3D spatio-temporal frequency domain.
An overview of the chapters follows.
Chapter 1 reviews the basics of multi-dimensional signals, transforms, and systems, which form the theoretical basis of many image and video processing methods.
We also address spatio-temporal sampling on MD lattices, which includes several
practical sampling structures such as progressive and interlaced sampling, as well
as theory of sampling structure conversion. Readers with a computer science background who may not be familiar with signal processing concepts can skip this chapter and start with Chapter 2.
Chapter 2 aims to provide a basic understanding of digital image and video
fundamentals. We cover the basic concepts of human vision, spatial frequency, color
models, analog and digital video representations, digital video standards, 3D stereo
and multi-view video representations, and evaluation of digital video quality. We
introduce popular digital video applications, including digital TV, digital cinema,
and video streaming over the Internet.
Chapter 3 addresses image (still-frame) filtering problems such as image
resampling (decimation and interpolation), gradient estimation and edge detection,
enhancement, de-noising, and restoration. Linear shift-invariant, adaptive, and nonlinear filters are considered. We provide a general framework for solution of ill-posed
inverse problems in Appendix B.
Chapter 4 covers 2D and 3D motion estimation methods. Motion estimation
is at the heart of digital video processing since motion is the most prominent feature of video, and motion-compensated filtering is the most effective way to utilize
temporal redundancy. Furthermore, many computer vision tasks require 2D or 3D
motion estimation and tracking as a first step. 2D motion estimation, which refers
to dense optical flow or sparse feature correspondence estimation, can be based on
nonparametric or parametric methods. Nonparametric methods include image gradient-based optical flow estimation, block matching, pel-recursive methods, Bayesian methods, and phase correlation methods. The parametric methods, based on the
affine model or the homography, can be used for image registration or to estimate
local deformations. 3D motion/structure estimation methods include those based
on the two-frame epipolar constraint (mainly for stereo pairs) or multi-frame factorization methods. Reconstruction of Euclidean 3D structure requires full-camera calibration while projective reconstruction can be performed without any calibration.
Chapter 5 introduces image segmentation and change detection, as well as segmentation of dominant motion or multiple motions using parameter clustering and
Bayesian methods. We also discuss simultaneous motion estimation and segmentation. Since two-view motion estimation techniques are very sensitive to inaccuracies
in the estimates of image gradients or point correspondences, motion tracking of
segmented objects over long monocular sequences or stereo pairs, which yield more
robust results, are also considered.
Chapter 6 addresses video filtering, including standards conversion, de-noising,
and super-resolution. It starts with the basic theory of motion-compensated filtering. Next, standards conversion problems, including frame rate conversion and
de-interlacing, are covered. Video frames often suffer from graininess, especially when
viewed in freeze-frame mode. Hence, motion-adaptive and motion-compensated
filtering for noise suppression are discussed. Finally, a comprehensive model for lowresolution video acquisition and super-resolution reconstruction methods (based on
this model) that unify various video filtering problems are presented.
Chapter 7 covers still-image, including binary (FAX) and gray-scale image, compression methods and standards such as JPEG and JPEG 2000. In particular, we
discuss lossless image compression and lossy discrete cosine transform coding and
wavelet coding methods.
Chapter 8 discusses video compression methods and standards that have made
digital video applications such as digital TV and digital cinema a reality. After a
brief introduction to different approaches to video compression, we cover MPEG-2,
AVC/H.264, and HEVC standards in detail, as well as their scalable video coding
and stereo/multi-view video coding extensions.
This textbook is the outcome of my experience in teaching digital image and
video processing for more than 20 years. It is comprehensive, written in a tutorial
style, which covers both fundamentals and the most recent progress in image filtering, motion estimation and tracking, image/video segmentation, video filtering, and
image/video compression with equal emphasis on these subjects. Unfortunately, it
is not possible to cover all state-of-the-art methods in digital video processing and
computer vision in a tutorial style in a single volume. Hence, only the most fundamental, popular techniques and algorithms are explained in a tutorial style. More
advanced algorithms and recent research results are briefly summarized and references are provided for self-study. Problem sets and MATLAB projects are included
at the end of each chapter for the reader to practice the methods.
Teaching materials will be provided to instructors upon request. A teaching plan
is provided in Table P.1, which assumes a 14-week semester with two 75-minute
classes each week, to cover the whole book in a one-semester digital image and video
processing course. Alternatively, it is possible to cover the book in two semesters,
which would allow time to delve into more technical details with each subject. The
first semester can be devoted to digital image processing, covering Chapters 1, 2, 3,
and 7. In the second semester, Chapters 4, 5, 6, and 8 can be covered in a follow-up
digital video processing course.
Clearly, this book is a compilation of knowledge collectively created by the signal
processing and computer science communities. I have included many citations and
references in each chapter, but I am sure I have neglected some since it is impossible
to give credit to all outstanding academic and industrial researchers who contributed to the development of image and video processing. Furthermore, outstanding
innovations in image and video coding are a result of work done by many scientists
Table P.1 Suggested Teaching Plan for a One-Semester Course
Lecture Topic
2D signals, 2D transforms
1.1, 1.2
2D systems, 2D FIR filters, frequency response
MD spatio-temporal sampling on lattices
1.4, 1.5
Digital images/video, human vision, video quality
Chapter 2
Vector-matrix notation, image models, formulation of ill-posed
problems in image/video processing
Appendix A,
Appendix B
Decimation, interpolation, multi-resolution pyramids
Gradient estimation, edge/corner detection
Image enhancement, point operations, unsharp masking, bilateral 3.1, 3.4
Noise filtering: LSI filters; adaptive, nonlinear, and non-local filters 3.5
Image restoration: iterative methods, POCS
Motion modeling, optical flow, correspondence
4.1, 4.2, 4.3
Differential methods: Lukas–Kanade, parametric models
Block matching, feature matching for parametric model
estimation, phase-correlation method
4.5, 4.7
3D motion estimation, epipolar geometry
Change detection, video segmentation
5.2, 5.3
Motion tracking
5.4, 5.5
Motion-compensated filtering, multi-frame de-interlacing,
6.1, 6.2, 6.3
Introduction to data/image compression, information theoretic
concepts, entropy coding, arithmetic coding
Lossless bitplane coding, group 3/4, JBIG standards
Predictive data coding, JPEG-LS standard
DCT and JPEG image compression
Wavelet transform, JPEG-2000 image compression
8.1, 8.2
MPEG-4 AVC/H.264 standard
Scalable video coding (SVC), DASH adaptive streaming,
3D/stereo and multi-view video compression
in various ISO and ITU groups over the years, where it is difficult to give individual
credit to everyone.
Finally, I would like to express my gratitude to Xin Li (WVU), Eli Saber, Moncef
Gabbouj, Janusz Konrad, and H. Joel Trussell for reviewing the manuscript at various stages. I would also like to thank Bernard Goodwin, Kim Boedigheimer, and
Julie Nahil from Prentice Hall for their help and support.
—A. Murat Tekalp
Koç University
Istanbul, Turkey
April 2015
This page intentionally left blank
About the Author
A. Murat Tekalp received a Ph.D. in electrical, computer, and systems engineering
from Rensselaer Polytechnic Institute (RPI), Troy, New York, in 1984. He was with
Eastman Kodak Company, Rochester, New York, from 1984 to 1987, and with the
University of Rochester, Rochester, New York, from 1987 to 2005, where he was
promoted to Distinguished University Professor. He is currently a professor at Koç
University, Istanbul, Turkey. He served as the Dean of Engineering at Koç University
from 2010 through 2013. His research interests are in the area of digital image and
video processing, image and video compression, and video networking.
Dr. Tekalp is a fellow of IEEE and a member of Academia Europaea and Turkish
Academy of Sciences. He received the TUBITAK Science Award (the highest scientific award in Turkey) in 2004. He is a former chair of the IEEE Technical Committee on Image and Multidimensional Signal Processing, and a founding member of
the IEEE Technical Committee on Multimedia Signal Processing. He was appointed
as the technical program co-chair for IEEE ICASSP 2000 in Istanbul, Turkey; the
general chair of IEEE International Conference on Image Processing (ICIP) at
Rochester, New York, in 2002; and technical program co-chair of EUSIPCO 2005
in Antalya, Turkey.
He was the editor-in-chief of the EURASIP journal Signal Processing: Image
Communication (published by Elsevier) from 2000 through 2010. He also served as
an associate editor for the IEEE Transactions on Signal Processing and IEEE Transactions on Image Processing. He was on the editorial board of IEEE’s Signal Processing
Magazine (2007–2010). He is currently on the editorial board of the Proceeedings
of the IEEE. He also serves as a member of the European Research Council (ERC)
Advanced Grant panels.
This page intentionally left blank
This page intentionally left blank
Digital Images and Video
Advances in ultra-high-definition and 3D-video technologies as well as high-speed
Internet and mobile computing have led to the introduction of new video services.
Digital images and video refer to 2D or 3D still and moving (time-varying) visual
information, respectively. A still image is a 2D/3D spatial distribution of intensity
that is constant with respect to time. A video is a 3D/4D spatio-temporal intensity pattern, i.e., a spatial-intensity pattern that varies with time. Another term
commonly used for video is image sequence, since a video is represented by a time
sequence of still images (pictures). The spatio-temporal intensity pattern of this time
sequence of images is ordered into a 1D analog or digital video signal as a function
of time only according to a progressive or interlaced scanning convention.
We begin with a short introduction to human visual perception and color models
in Section 2.1. We give a brief review of analog-video representations in Section 2.2,
mainly to provide a historical perspective. Next, we present 2D digital video representations and a brief summary of current standards in Section 2.3. We introduce
3D digital video display, representations, and standards in Section 2.4. Section 2.5
provides an overview of popular digital video applications, including digital TV,
digital cinema, and video streaming. Finally, Section 2.6 discusses factors affecting
video quality and quantitative and subjective video-quality assessment.
Chapter 2. Digital Images and Video
2.1 Human Visual System and Color
Video is mainly consumed by the human eye. Hence, many imaging system design
choices and parameters, including spatial and temporal resolution as well as color
representation, have been inspired by or selected to imitate the properties of human
vision. Furthermore, digital image/video-processing operations, including filtering
and compression, are generally designed and optimized according to the specifications of the human eye. In most cases, details that cannot be perceived by the human
eye are regarded as irrelevant and referred to as perceptual redundancy.
2.1.1 Color Vision and Models
The human eye is sensitive to the range of wavelengths between 380 nm (blue end
of the visible spectrum) and 780 nm (red end of the visible spectrum). The cornea,
iris, and lens comprise an optical system that forms images on the retinal surface.
There are about 100-120 million rods and 7-8 million cones in the retina [Wan
95, Fer 01]. They are receptor nerve cells that emit electrical signals when light hits
them. The region of the retina with the highest density of photoreceptors is called
the fovea. Rods are sensitive to low-light (scotopic) levels but only sense the intensity
of the light; they enable night vision. Cones enable color perception and are best in
bright (photopic) light. They have bandpass spectral response. There are three types
of cones that are more sensitive to short (S), medium (M), and long (L) wavelengths,
respectively. The spectral response of S-cones peak at 420 nm, M-cones at 534 nm,
and L-cones at 564 nm, with significant overlap in their spectral response ranges and
varying degrees of sensitivity at these range of wavelengths specified by the function
mk (), k 5 r, g, b, as depicted in Figure 2.1(a).
The perceived color of light f (x1, x2, ) at spatial location (x1, x2) depends on the
distribution of energy in the wavelength l dimension. Hence, color sensation can
be achieved by sampling l into three levels to emulate color sensation of each type
of cones as:
f k ( x1 , x 2 ) ∫ f ( x1 , x 2 , ) mk ( )d k r , g , b (2.1)
where mk () is the wavelength sensitivity function (also known as the colormatching function) of the kth cone type or color sensor. This implies that perceived
color at any location (x1, x2) depends only on three values fr , fg, and fb, which are
called the tristimulus values.
It is also known that the human eye has a secondary processing stage whereby
the R, G, and B values sensed by the cones are converted into a luminance and two
2.1 Human Visual System and Color
color-difference (chrominance) values [Fer 01]. The luminance Y is related to the
perceived brightness of the light and is given by
Y ( x1 , x 2 ) ∫ f ( x1 , x 2 , ) l ( ) d (2.2)
where l(l) is the International Commission on Illumination (CIE) luminous efficiency function, depicted in Figure 2.1(b), which shows the contribution of energy
at each wavelength to a standard human observer’s perception of brightness. Two
chrominance values describe the perceived color of the light. Color representations
for color image processing are further discussed in Section 2.3.3.
Relative Response
Wavelength (nm)
Luminous Efficiency
550 600 650 700
Wavelength (nm)
Figure 2.1 Spectral sensitivity: (a) CIE 1931 color-matching functions for a standard observer with
x, −
y , and −
z may represent mr (l), mg (l), and mb (l),
a 2-degree field of view, where the curves −
respectively, and (b) the CIE luminous efficiency function l(l) as a function of wavelength l.
Chapter 2. Digital Images and Video
Now that we have established that the human eye perceives color in terms of
three component values, the next question is whether all colors can be reproduced
by mixing three primary colors. The answer to this question is yes in the sense that
most colors can be realized by mixing three properly chosen primary colors. Hence,
inspired by human color perception, digital representation of color is based on the
tri-stimulus theory, which states that all colors can be approximated by mixing
three additive primaries, which are described by their color-matching functions. As
a result, colors are represented by triplets of numbers, which describe the weights
used in mixing the three primaries. All colors that can be reproduced by a combination of three primary colors define the color gamut of a specific device. There
are different choices for selecting primaries based on additive and subtractive color
models. We discuss the additive RGB and subtractive CMYK color spaces and color
management in the following. However, an in-depth discussion of color science is
beyond the scope of this book, and interested readers are referred to [Tru 93, Sha 98,
Dub 10].
RGB and CMYK Color Spaces
The RGB model, inspired by human vision, is an additive color model in which red,
green, and blue light are added together to reproduce a variety of colors. The RGB
model applies to devices that capture and emit color light such as digital cameras,
video projectors, LCD/LED TV and computer monitors, and mobile phone displays. Alternatively, devices that produce materials that reflect light, such as color
printers, are governed by the subtractive CMYK (Cyan, Magenta, Yellow, Black)
color model. Additive and subtractive color spaces are depicted in Figure 2.2. RGB
and CMYK are device-dependent color models: i.e., different devices detect or reproduce a given RGB value differently, since the response of color elements (such as
filters or dyes) to individual R, G, and B levels may vary among different manufacturers. Therefore, the RGB color model itself does not define absolute red, green, and
blue (hence, the result of mixing them) colorimetrically.
When the exact chromaticities of red, green, and blue primaries are defined,
we have a color space. There are several color spaces, such as CIERGB, CIEXYZ, or
sRGB. CIERGB and CIEXYZ are the first formal color spaces defined by the CIE
in 1931. Since display devices can only generate non-negative primaries, and an
adequate amount of luminance is required, there is, in practice, a limitiation on the
gamut of colors that can be reproduced on a given device. Color characteristics of a
device can be specified by its International Color Consortium (ICC) profile.
2.1 Human Visual System and Color
Figure 2.2 Color spaces: (a) additive color space and (b) subtractive color space.
Color Management
Color management must be employed to generate the exact same color on different
devices, where the device-dependent color values of the input device, given its ICC profile, is first mapped to a standard device-independent color space, sometimes called the
Profile Connection Space (PCS), such as CIEXYZ. They are then mapped to the devicedependent color values of the output device given the ICC profile of the output device.
Hence, an ICC profile is essentially a mapping from a device color space to the PCS
and from the PCS to a device color space. Suppose we have particular RGB and CMYK
devices and want to convert the RGB values to CMYK. The first step is to obtain the
ICC profiles of concerned devices. To perform the conversion, each (R, G, B) triplet is
first converted to the PCS using the ICC profile of the RGB device. Then, the PCS is
converted to the C, M, Y, and K values using the profile of the second device.
Color management may be side-stepped by calibrating all devices to a common
standard color space, such as sRGB, which was developed by HP and Microsoft
in 1996. sRGB uses the color primaries defined by the ITU-R recommendation
BT.709, which standardizes the format of high-definition television. When such a
calibration is done well, no color translations are needed to get all devices to handle
colors consistently. Avoiding the complexity of color management was one of the
goals in developing sRGB [IEC 00].
2.1.2 Contrast Sensitivity
Contrast can be defined as the difference between the luminance of a region and its
background. The human visual system is more sensitive to contrast than absolute
Chapter 2. Digital Images and Video
luminance; hence, we can perceive the world around us similarly regardless of changes
in illumination. Since most images are viewed by humans, it is important to understand how the human visual system senses contrast so that algorithms can be designed
to preserve the more visible information and discard the less visible ones. Contrastsensitivity mechanisms of human vision also determine which compression or processing artifacts we see and which we don’t. The ability of the eye to discriminate
between changes in intensity at a given intensity level is quantified by Weber’s law.
Weber’s Law
Weber’s law states that smaller intensity differences are more visible on a darker background and can be quantified as
c (constant), for I 0 I
where DI is the just noticeable difference (JND) [Gon 07]. Eqn. (2.5) states that the
JND grows proportional to the intensity level I. Note that I 5 0 denotes the darkest intensity, while I 5 255 is the brightest. The value of c is empirically found to be
around 0.02. The experimental set-up to measure the JND is shown in Figure 2.3(a).
The rods and cones comply with Weber’s law above -2.6 log candelas (cd)/m2 (moonlight) and 2 log cd/m2 (indoor) luminance levels, respectively [Fer 01].
Brightness Adaptation
The human eye can adapt to different illumination/intensity levels [Fer 01]. It has
been observed that when the background-intensity level the observer has adapted to
is different from I, the observer’s intensity resolution ability decreases. That is, when
I0 is different from I, as shown in Figure 2.3(b), the JND DI increases relative to
the case I0 5 I. Furthermore, the simultanenous contrast effect illustrates that humans
perceive the brightness of a square with constant intensity differently as the intensity
of the background varies from light to dark [Gon 07].
It is also well-known that the human visual system undershoots and overshoots
around the boundary of step transitions in intensity as demonstrated by the Mach
band effect [Gon 07].
Visual Masking
Visual masking refers to a nonlinear phenomenon experimentally observed in the
human visual system when two or more visual stimuli that are closely coupled in
space or time are presented to a viewer. The action of one visual stimulus on the
visibility of another is called masking. The effect of masking may be a decrease in
2.1 Human Visual System and Color
I∆ I
I∆ I
I∆ I I
I I∆ I
Figure 2.3 Illustration of (a) the just noticeable difference and (b) brightness adaptation.
brightness or failure to detect the target or some details, e.g., texture. Visual masking
can be studied under two cases: spatial masking and temporal masking.
Spatial Masking
Spatial masking is observed when a viewer is presented with a superposition of a
target pattern and mask (background) image [Fer 01]. The effect states that the visibility of the target pattern is lower when the background is spatially busy. Spatial busyness measures include local image variance or textureness. Spatial masking
implies that visibility of noise or artifact patterns is lower in spatially busy areas of an
image as compared to spatially uniform image areas.
Temporal Masking
Temporal masking is observed when two stimuli are presented sequentially [Bre 07].
Salient local changes in luminance, hue, shape, or size may become undetectable in the
presence of large coherent object motion [Suc 11]. Considering video frames as a sequence
of stimuli, fast-moving objects and scene cuts can trigger a temporal-masking effect.
2.1.3 Spatio-Temporal Frequency Response
An understanding of the response of the human visual system to spatial and temporal frequencies is important to determine video-system design parameters and video-­
compression parameters, since frequencies that are invisible to the human eye are
Chapter 2. Digital Images and Video
Spatial-Frequency Response
Spatial frequencies are related to how still (static) image patterns vary in the horizontal
and vertical directions in the spatial plane. The spatial-frequency response of the human
eye varies with the viewing distance; i.e., the closer we get to the screen the better we can
see details. In order to specify the spatial frequency independent of the viewing distance,
spatial frequency (in cycles/distance) must be normalized by the viewing distance d,
which can be done by defining the viewing angle  as shown in Figure 2.4(a).
w /2
, considering
Let w denote the picture width. If w /2  d , then  sin 2
the right triangle formed by the viewer location, an end of the picture, and the
middle of the picture. Hence,
(radians) =
(degrees) d
Let fw denote the number of cycles per picture width, then the normalized horizontal spatial frequency (i.e., number of cycles per viewing degree) fu is given by
f fw
f d
d f w
w (cycles / radian) (cycles / degree) w
180 w
The normalized vertical spatial frequency can be defined similarly in the units of
cycles/degree. As we move away from the screen d increases, and the same number of
cycles per picture width fw appears as a larger frequency fu per viewing degree. Since
the human eye has reduced contrast sensitivity at higher frequencies, the same pattern is more difficult to see from a larger distance d. The horizontal and vertical resolution (number of pixels and lines) of a TV has been determined such that horizontal
and vertical sampling frequencies are twice the highest frequency we can see (according to the Nyquist sampling theorem), assuming a fixed value for the ratio d/w—i.e.,
viewing distance over picture width. Given a fixed viewing distance, clearly we need
more video resolution (pixels and lines) as picture (screen) size increases to experience the same video quality.
Figure 2.4(b) shows the spatial-frequency response, which varies by the average
luminance level, of the eye for both the luminance and chrominance components
of still images. We see that the spatial-frequency response of the eye, in general, has
low-pass/band-pass characteristics, and our eyes are more sensitive to higher frequency patterns in the luminance components compared with those in the chrominance components. The latter observation is the basis of the conversion from RGB
to the luminance-chrominance space for color image processing and the reason we
subsample the two chrominance components in color image/video compression.
2.1 Human Visual System and Color
0.03 0.1
Spatial frequency (cycles/degree)
Figure 2.4 Spatial frequency and spatial response: (a) viewing angle
and (b) spatial-frequency response of the human eye [Mul 85].
Temporal-Frequency Response
Video is displayed as a sequence of still frames. The frame rate is measured in terms of
the number of pictures (frames) displayed per second or Hertz (Hz). The frame rates
for cinema, television, and computer monitors have been determined according to
the temporal-frequency response of our eyes. The human eye has lower sensitivity to
higher temporal frequencies due to temporal integration of incoming light into the
retina, which is also known as vision persistence. It is well known that the integration
period is inversely proportional to the incoming light intensity. Therefore, we can see
higher temporal frequencies on brighter screens. Psycho-visual experiments indicate
the human eye cannot perceive flicker if the refresh rate of the display (temporal frequency) is more than 50 times per second for TV screens. Therefore, the frame rate
for TV is set at 50-60 Hz, while the frame rate for brighter computer monitors is 72
Hz or higher, since the brighter the screen the higher the critical flicker frequency.
Interaction Between Spatial- and Temporal-Frequency Response
Video exhibits both spatial and temporal variations, and spatial- and temporalfrequency responses of the eye are not mutually independent. Hence, we need to
understand the spatio-temporal frequency response of the eye. The effects of changing average luminance on the contrast sensitivity for different combinations of spatial
and temporal frequencies have been investigated [Nes 67]. Psycho-visual experiments
Chapter 2. Digital Images and Video
indicate that when the temporal (spatial) frequencies are close to zero, the spatial
(temporal) frequency response has bandpass characteristics. At high temporal (spatial) frequencies, the spatial (temporal) frequency response has low-pass characteristics with smaller cut-off frequency as temporal (spatial) frequency increases. This
implies that we can exchange spatial video resolution for temporal resolution, and
vice versa. Hence, when a video has high motion (moves fast), the eyes cannot sense
high spatial frequencies (details) well if we exclude the effect of eye movements.
Eye Movements
The human eye is similar to a sphere that is free to move like a ball in a socket. If
we look at a nearby object, the two eyes turn in; if we look to the left, the right eye
turns in and the left eye turns out; if we look up or down, both eyes turn up or down
together. These movements are directed by the brain [Hub 88]. There are two main
types of gaze-shifting eye movements, saccadic and smooth pursuit, that affect the
spatial- and spatio-temporal frequency response of the eye. Saccades are rapid movements of the eyes while scanning a visual scene. “Saccadic eye movements” enable
us to scan a greater area of the visual scene with the high-resolution fovea of the eye.
On the other hand, “smooth pursuit” refers to movements of the eye while tracking
a moving object, so that a moving image remains nearly static on the high-resolution
fovea. Obviously, smooth pursuit eye movements affect the spatio-temporal frequency response of the eye. This effect can be modeled by tracking eye movements
of the viewer and motion compensating the contrast sensitivity function accordingly.
2.1.4 Stereo/Depth Perception
Stereoscopy creates the illusion of 3D depth from two 2D images, a left and a right
image that we should view with our left and right eyes. The horizontal distance
between the eyes (called interpupilar distance) of an average human is 6.5 cm. The
difference between the left and right retinal images is called binocular disparity. Our
brain deducts depth information from this binocular disparity. 3D display technologies that enable viewing of right and left images with our right and left eyes, respectively, are discussed in Section 2.4.1.
Accomodation, Vergence, and Visual Discomfort
In human stereo vision, there are two oculomotor mechanisms, accommodation
(where we focus) and vergence (where we look), which are reflex eye movements.
Accommodation is the process by which the eye changes optical focus to maintain a
clear image of an object as its distance from the eye varies. Vergence or convergence
2.2 Analog Video
are the movements of both eyes to make sure the image of the object being looked at
falls on the corresponding spot on both retinas. In real 3D vision, accommodation
and vergence distances are the same. However, in flat 3D displays both left and right
images are displayed on the plane of the screen, which determines the accommodation distance, while we look and perceive 3D objects at a different distance (usually
closer to us), which is the vergence distance. This difference between accommodation and vergence distances may cause serious discomfort if it is greater than some
tolerable amount. The depth of an object in the scene is determined by the disparity
value, which is the displacement of a feature point between the right and left views.
The depth, hence the difference between accommodation and vergence distances,
can be controlled by 3D-video (disparity) processing at the content preparation stage
to provide a comfortable 3D viewing experience.
Another cause of viewing discomfort is the cross-talk between the left and right
views, which may cause ghosting and blurring. Cross-talk may result from imperfections in polarizing filters (passive glasses) or synchronization errors (active shutters),
but it is more prominent in auto-stereoscopic displays where the optics may not
completely prevent cross-talk between the left and right views.
Binocular Rivalry/Suppression Theory
Binocular rivalry is a visual perception phenomenon that is observed when different
images are presented to right and left eyes [Wad 96]. When the quality difference
between the right and left views are small, according to the suppression theory of
stereo vision, the human eye can tolerate absence of high-frequency content in one
of the views; therefore, two views can be represented at unequal spatial resolutions
or quality. This effect has lead to asymmetric stereo-video coding, where only the
dominant view is encoded with high fidelity (bitrate). The results have shown that
perceived 3D-video quality of such asymmetric processed stereo pairs is similar to
that of symmetrically encoded sequences at higher total bitrate. They also observe
that scaling (zoom in/out) one or both views of a stereoscopic test sequence does not
affect depth perception. We note that these results have been confirmed on short test
sequences. It is not known whether asymmetric view resolution or quality would
cause viewing discomfort over longer videos with increased period of viewing.
2.2 Analog Video
We used to live in a world of analog images and video, where we dealt with photographic film, analog TV sets, videocassette recorders (VCRs), and camcorders.
Chapter 2. Digital Images and Video
For video distribution, we relied on analog TV broadcasts and analog cable TV,
which transmitted predetermined programming at a fixed rate. Analog video, due
to its nature, provided a very limited amount of interactivity, e.g., only channel
selection on the TV and fast-forward search and slow-motion replay on the VCR.
Additionally, we had to live with the NTSC/PAL/SECAM analog signal formats
with their well-known artifacts and very low still-frame image quality. In order
to display NTSC signals on computer monitors or European TV sets, we needed
expensive transcoders. In order to display a smaller version of the NTSC picture
in a corner of the monitor, we first had to digitize the whole picture and then
digitally reduce its size. Searching a video archive for particular footage required
tedious visual scanning of a whole bunch of videotapes. Motion pictures were
recorded on photographic film, which is a high-resolution analog medium, or
on laser discs as analog signals using optical technology. Manipulation of analog
video is not an easy task, since it requires digitization of the analog signal into
digital form first.
Today almost all video capture, processing, transmission, storage, and search are
in digital form. In this section, we describe the nature of the analog-video signal
because an understanding of history of video and the limitations of analog video
formats is important. For example, interlaced scanning originates from the history
of analog video. We note that video digitized from analog sources is limited by the
resolution and the artifacts of the respective analog signal.
2.2.1 Progressive vs. Interlaced Scanning
The analog-video signal refers to a one-dimensional (1D) signal s(t) of time that is
obtained by sampling sc(x1, x2, t) in the vertical x2 and temporal coordinates. This
conversion of 3D spatio-temporal signal into a 1D temporal signal by periodic
vertical-temporal sampling is called scanning. The signal s(t), then, captures the
time-varying image intensity sc(x1, x2, t) only along the scan lines. It also contains the
timing information and blanking signals needed to align pictures.
The most commonly used scanning methods are progressive scanning and interlaced scanning. Progressive scan traces a complete picture, called a frame, at every
Dt sec. The spot flies back from B to C, called the horizontal retrace, and from D to
A, called the vertical retrace, as shown in Figure 2.5(a). For example, the computer
industry uses progressive scanning with Dt51/72 sec for monitors. On the other
hand, the TV industry uses 2:1 interlaced scan where the odd-numbered and evennumbered lines, called the odd field and the even field, respectively, are traced in
turn. A 2:1 interlaced scanning raster is shown in Figure 2.5(b), where the solid line
2.2 Analog Video
Figure 2.5 Scanning raster: (a) progressive scan; (b) interlaced scan.
and the dotted line represent the odd and the even fields, respectively. The spot snaps
back from D to E, and from F to A, for even and odd fields, respectively, during the
vertical retrace intervals.
2.2.2 Analog-Video Signal Formats
Some important parameters of the video signal are the vertical resolution, aspect
ratio, and frame/field rate. The vertical resolution is related to the number of scan
lines per frame. The aspect ratio is the ratio of the width to the height of a frame.
As discussed in Section 2.1.3, the human eye does not perceive flicker if the refresh
rate of the display is more than 50 Hz. However, for analog TV systems, such a
high frame rate, while preserving the vertical resolution, requires a large transmission bandwidth. Thus, it was determined that analog TV systems should use interlaced scanning, which trades vertical resolution to reduced flickering within a fixed
An example analog-video signal s(t) is shown in Figure 2.6. Blanking pulses
(black) are inserted during the retrace intervals to blank out retrace lines on the monitor. Sync pulses are added on top of the blanking pulses to synchronize the receiver’s
horizontal and vertical sweep circuits. The sync pulses ensure that the picture starts
at the top-left corner of the receiving monitor. The timing of the sync pulses is, of
course, different for progressive and interlaced video.
Several analog-video signal standards, which are obsolete today, have different
image parameters (e.g., spatial and temporal resolution) and differ in the way they
handle color. These can be grouped as: i) component analog video; ii) composite
video; and iii) S-video (Y/C video). Component analog video refers to individual
Chapter 2. Digital Images and Video
sync pulse
Active line time
Horizontal retrace
t, µs
Figure 2.6 Analog-video signal for one full line.
red (R), green (G), and blue (B) video signals. Composite-video format encodes the
chrominance components on top of the luminance signal for distribution as a single
signal that has the same bandwidth as the luminance signal. Different compositevideo formats, e.g., NTSC (National Television Systems Committee), PAL (Phase
Alternation Line), and SECAM (Systeme Electronique Color Avec Memoire), have
been used in different regions of the world. The composite signal usually results in
errors in color rendition, known as hue and saturation errors, because of inaccuracies
in the separation of the color signals. S-video is a compromise between the composite
video and component video, where we represent the video with two component signals, a luminance and a composite chrominance signal. The chrominance signals have
been based on (I,Q) or (U,V) representation for NTSC, PAL, or SECAM systems.
S-video was used in consumer-quality videocasette recorders and analog camcorders
to obtain image quality better than that of composite video. Cameras specifically
designed for analog television pickup from motion picture film were called telecine
cameras. They employed frame-rate conversion from 24 frames/sec to 60 fields/sec.
2.2.3 Analog-to-Digital Conversion
The analog-to-digital (A/D) conversion process consists of pre-filtering (for antialiasing), sampling, and quantization of component (R, G, B) signal or composite
signal. The ITU (International Telecommunications Union) and SMPTE (Society
of Motion Picture and Television Engineers) have standardized sampling parameters
for both component and composite video to enable easy exchange of digital video
2.3 Digital Video
across different platforms. For A/D conversion of component signals, the horizontal
sampling rate of 13.5 MHz for the luma component and 6.75 MHz for two chroma
components were chosen, because they satisfy the following requirements:
1. Minimum sampling frequency (Nyquist rate) should be 4.2 3 2 5 8.4 MHz for
525/30 NTSC luma and 5 3 2 5 10 MHz for 625/50 PAL luma signals.
2. Sampling rate should be an integral multiple of the line rate, so samples in successive lines are correctly aligned (on top of each other).
3. For sampling component signals, there should be a single rate for 525/30 and
625/50 systems; i.e., the sampling rate should be an integral multiple of line
rates (lines/sec) of both 29.97 3 525 5 15,734 and 25 3 625 5 15,625.
For sampling the composite signal, the sampling frequency must be an integral
multiple of the sub-carrier frequency to simplify composite signal to RGB decoding
of sampled signal. It is possible to operate at 3 or 4 times the subcarrier frequency,
although most systems choose to employ 4 3 3.58 5 14.32 MHz for NTSC and
4 3 4.43 5 17.72 MHz for PAL signals, respectively.
2.3 Digital Video
We have experienced a digital media revolution in the last couple of decades. TV
and cinema have gone all-digital and high-definition, and most movies and some TV
broadcasts are now in 3D format. High-definition digital video has landed on laptops, tablets, and cellular phones with high-quality media streaming over the Internet. Apart from the more robust form of the digital signal, the main advantage of
digital representation and transmission is that they make it easier to provide a diverse
range of services over the same network. Digital video brings broadcasting, cinema,
computers, and communications industries together in a truly revolutionary manner, where telephone, cable TV, and Internet service providers have become fierce
competitors. A single device can serve as a personal computer, a high-definition TV,
and a videophone. We can now capture live video on a mobile device, apply digital
processing on a laptop or tablet, and/or print still frames at a local printer. Other
applications of digital video include medical imaging, surveillance for military and
law enforcement, and intelligent highway systems.
2.3.1 Spatial Resolution and Frame Rate
Digital-video systems use component color representation. Digital color cameras
provide individual RGB component outputs. Component color video avoids the
Chapter 2. Digital Images and Video
artifacts that result from analog composite encoding. In digital video, there is no
need for blanking or sync pulses, since it is clear where a new line starts given the
number of pixels per line.
The horizontal and vertical resolution of digital video is related to the pixel sampling density, i.e., the number of pixels per unit distance. The number of pixels per
line and the number of lines per frame is used to classify video as standard, high, or
ultra-high definition, as depicted in Figure 2.7. In low-resolution digital video, pixellation (aliasing) artifact arises due to lack of sufficient spatial resolution. It manifests
itself as jagged edges resulting from individual pixels becoming visible. The visibility
of pixellation artifacts varies with the size of the display and the viewing distance.
This is quite different from analog video where the lack of spatial-resolution results
in blurring of image in the respective direction.
The frame/field rate is typically 50/60 Hz, although some displays use frame interpolation to display at 100/120, 200 or even 400 Hz. The notation 50i (or 60i) indicates interlaced video with 50 (60) fields/sec, which corresponds to 25 (30) pictures/
sec obtained by weaving the two fields together. On the other hand, 50p (60p) denotes
50 (60) full progressive frames/sec.
The arrangement of pixels and lines in a contiguous region of the memory is
called a bitmap. There are five key parameters of a bitmap: the starting address in
the memory, the number of pixels per line, the pitch value, the number of lines,
and the number of bits per pixel. The pitch value specifies the distance in memory
from the start of one line to the next. The most common use of pitch different from
Ultra HD
3840 x 2160
HD 1280 x 720
Full HD
1920 x 1080
720 x 576
720 x 488
Figure 2.7 Digital-video spatial-resolution formats.
2.3 Digital Video
the number of pixels per line is to set pitch to the next highest power of 2, which
may help certain applications run faster. Also, when dealing with interlaced inputs,
setting the pitch to double the number of pixels per line facilitates writing lines from
each field alternately in memory. This will form a “weaved frame” in a contiguous
region of the memory.
2.3.2 Color, Dynamic Range, and Bit-Depth
This section addresses color representation, dynamic range, and bit-depth in digital
Color Capture and Display
Color cameras can be the three-sensor type or single-sensor type. Three-sensor cameras capture R, G, and B components using different CCD panels, using an optical beam splitter; however, they may suffer from synchronicity problems and high
cost, while single-sensor cameras often have to compromise spatial resolution. This is
because a color filter array is used so that each CCD element captures one of R, G, or
B pixels in some periodic pattern. A commonly used color filter pattern is the Bayer
array, shown in Figure 2.8, where two out of every four pixels are green, one is red,
and one is blue, since green signal contributes the most to the luminance channel.
The missing pixel values in each color channel are computed by linear or adaptive
Figure 2.8 Bayer color-filter array pattern.
Chapter 2. Digital Images and Video
interpolation filters, which may result in some aliasing artifacts. Similar color filter
array patterns are also employed in LCD/LED displays, where the human eye performs low-pass filtering to perceive a full-colored image.
Dynamic Range
The dynamic range of a capture device (e.g., a camera or scanner) or a display device
is the ratio between the maximum and minimum light intensities that can be represented. The luminance levels in the environment range from 24 log cd/m2 (starlight) to 6 log cd/m2 (sun light); i.e., the dynamic range is about 10 log units [Fer
01]. The human eye has complex fast and slow adaptation schemes to cope with this
large dynamic range. However, a typical imaging device (camera or display) has a
maximum dynamic range of 300:1, which corresponds to 2.5 log units. Hence, our
ability to capture and display a foreground object subject to strong backlighting with
proper contrast is limited. High dynamic range (HDR) imaging aims to remedy this
HDR Image Capture
HDR image capture with a standard dynamic range camera requires taking a
sequence of pictures at different exposure levels, where raw pixel exposure data (linear in exposure time) are combined by weighted averaging to obtain a single HDR
image [Gra 10]. There are two possible ways to display HDR images: i) employ
new higher dynamic range display technologies, or ii) employ local tone-mapping
algorithms for dynamic range compression (see Chapter 3) to better render details in
bright or dark areas on a standard display [Rei 07].
HDR Displays
Recently, new display technologies that are capable of up to 50,000:1 or 4.7 log units
dynamic range with maximum intensity 8500 cd/m2, compared to standard displays
with contrast ratio 2 log units and maximum intensity 300 cd/m2, have been proposed [See 04]. This high dynamic range matches the human eye’s short time-scale
(fast) adaptation capability well, which enables our eyes to capture approximately 5
log units of dynamic range at the same time.
Image-intensity values at each sample are quantized for a finite-precision representation. Today, each color component signal is typically represented with 8 bits per
pixel, which can capture 255:1 dynamic range for a total of 24 bits/pixel and 224
2.3 Digital Video
distinct colors to avoid “contouring artifacts.” Contouring results in slowly varying
regions of image intensity due to insufficient bit resolution. Some applications, such
as medical imaging and post-production editing of motion pictures may require 10,
12, or more bits/pixel/color. In high dynamic range imaging, 16 bits/pixel/color is
required to capture a 50,000:1 dynamic range, which is now supported in JPEG.
Digital video requires much higher data rates and transmission bandwidths as
compared to digital audio. CD-quality digital audio is represented with 16 bits/
sample, and the required sampling rate is 44 kHz. Thus, the resulting data rate is
approximately 700 kbits/sec (kbps). This is multiplied by 2 for stereo audio. In comparison, a high-definition TV signal has 1920 pixels/line and 1080 lines for each
luminance frame, and 960 pixels/line and 540 lines for each chrominance frame.
Since we have 25 frames/sec and 8 bits/pixel/color, the resulting data rate exceeds
700 Mbps, which testifies to the statement that a picture is worth 1000 words! Thus,
the feasibility of digital video is dependent on image-compression technology.
2.3.3 Color Image Processing
Color images/video are captured and displayed in the RGB format. However, they
are often converted to an intermediate representation for efficient compression and
processing. We review the luminance-chrominance (for compression and filtering)
and the normalized RGB and hue-saturation-intensity (HSI) (for color-specific processing) representations in the following.
The luminance-chrominance color model was used to develop an analog color TV
transmission system that is backwards compatible with the legacy analog black and
white TV systems. The luminance component, denoted by Y, corresponds to the
gray-level representation of video, while the two chrominance components, denoted
by U and V for analog video or Cr and Cb for digital video, represent the deviation
of color from the gray level on blue–yellow and red–cyan axes. It has been observed
that the human visual system is less sensitive to variations (higher frequencies) in
chrominance components (see Figure 2.4(b)). This has resulted in the subsampled
chrominance formats, such as 4:2:2 and 4:2:0. In the 4:2.2 format, the chrominance components are subsampled only in the horizontal direction, while in 4:2:0
they are subsampled in both directions as illustrated in Figure 2.9. The luminancechrominance representation offers higher compression efficiency, compared to the
RGB representation due to this subsampling.
Chapter 2. Digital Images and Video
Figure 2.9 Chrominance subsampling formats: (a) no subsampling; (b) 4:2:2; (c) 4:2:0 format.
ITU-R BT.709 defines the conversion between RGB and YCrCb representations as:
Y 0.299 R 0.587 G 0.114 B
Cr 0.499 R 0.418 G 0.0813 B 128 Cb 0.169 R 0.331 G 0.499 B 128
which states that the human visual system perceives the contribution of R-G-B to
image intensity approximately with a 3-6-1 ratio, i.e., red is weighted by 0.3, green
by 0.6 and blue by 0.1.
The inverse conversion is given by
R Y 1.402 (Cr 128)
G Y 0.714 (Cr 128) 0.344 (Cb 128) B Y 1.772 (Cb 128)
The resulting R, G, and B values must be truncated to the range (0, 255) if they fall
outside. We note that Y-Cr-Cb is not a color space. It is a way of encoding the RGB
information, and actual colors displayed depends on the specific RGB space used.
A common practice in color image processing, such as edge detection, enhancement, denoising, restoration, etc., in the luminance-chrominance domain is to process only the luminance (Y) component of the image. There are two main reasons
for this: i) processing R, G, and B components independently may alter the color
balance of the image, and ii) the human visual system is not very sensitive to high frequencies in the chrominance components. Therefore, we first convert a color image
2.3 Digital Video
into Y-Cr-Cb color space, then perform image enhancement, denoising, restoration,
etc., on the Y channel only. We then transform the processed Y channel and unprocessed Cr and Cb channels back to the R-G-B domain for display.
Normalized rgb
Normalized rgb components aim to reduce the dependency of color represented by
the RGB values on image brightness. They are defined by
r = R / (R + G + B )
g = G / (R + G + B ) (2.8)
b = B / (R + G + B )
The normalized r, g, b values are always within the range 0 to 1, and
r g b 1 (2.9)
Hence, they can be specified by any two components, typically by (r, g) and the third
component can be obtained from Eqn. (2.9). The normalized rgb domain is often
used in color-based object detection, such as skin-color or face detection.
Example. We demonstrate how the normalized rgb domain helps to
detect similar colors independent of brightness by means of an example:
Let’s assume we have two pixels with (R, G, B) values (230, 180, 50) and
(115, 90, 25). It is clear that the second pixel is half as bright as the first,
which may be because it is in a shadow. In the normalized rgb, both pixels
are represented by r 5 0.50, g 5 0.39, and b 5 0.11. Hence, it is apparent
that they represent the same color after correcting for brightness difference
by the normalization.
Hue-Saturation-Intensity (HSI)
Color features that best correlate with human perception of color are hue, saturation, and intensity. Hue relates to the dominant wavelength, saturation relates to the
spread of power about this wavelength (purity of the color), and intensity relates to
the perceived luminance (similar to the Y channel). There is a family of color spaces
that specify colors in terms of hue, saturation, and intensity, known as HSI spaces.
Conversion to HSI where each component is in the range [0,1] can be performed
from the scaled RGB, where each component is divided by 255 so they are in the
Chapter 2. Digital Images and Video
range [0,1]. The HSI space specifies color in cylindrical coordinates and conversion
formulas (2.10) are nonlinear [Gon 07].
 1 [( R G ) ( R B )]  
if B G where arccos 
H  
 360 if B G
 ( R G )2 ( R B )(G B ) 
S 1
3min{ R ,G , B }
R G B (2.10)
Note that HSI is not a perceptually uniform color space, i.e., equal perturbations
in the component values do not result in perceptually equal color variations across
the range of component values. The CIE has also standardized some perceptually
uniform color spaces, such as L*, u*, v* and L*, a*, b* (CIELAB).
2.3.4 Digital-Video Standards
Exchange of digital video between different products, devices, and applications requires
digital-video standards. We can group digital-video standards as video-format (resolution) standards, video-interface standards, and image/video compression standards. In
the early days of analog TV, cinema (film), and cameras (cassette), the computer, TV,
and consumer electronics industries established different display resolutions and scanning standards. Because digital video has brought cinema, TV, consumer electronics,
and computer industries ever closer, standardization across industries has started. This
section introduces recent standards and standardization efforts.
Video-Format Standards
Historically, standardization of digital-video formats originated from different
sources: ITU-R driven by the TV industry, SMPTE driven by the motion picture
industry, and computer/consumer electronics associations.
Digital video was in use in broadcast TV studios even in the days of analog TV,
where editing and special effects were performed on digitized video because it is easier
to manipulate digital images. Working with digital video avoids artifacts that would
otherwise be caused by repeated analog recording of video on tapes during various
production stages. Digitization of analog video has also been needed for conversion
2.3 Digital Video
between different analog standards, such as from PAL to NTSC, and vice versa.
ITU-R (formerly CCIR) Recommendation BT.601 defines a standard definition TV
(SDTV) digital-video format for 525-line and 625-line TV systems, also known
as digital studio standard, which is originally intended to digitize analog TV signals to permit digital post-processing as well as international exchange of programs.
This recommendation is based on component video with one luminance (Y) and
two chrominance (Cr and Cb) signals. The sampling frequency for analog-to-digital
(A/D) conversion is selected to be an integer multiple of the horizontal sweep frequencies (line rates) fh,525 5 525 5 29.97 5 15,734 and fh,625 5 625 3 25 5 15,625
in both 525- and 625-line systems, which was discussed in Section 2.2.3. Thus, for
the luminance
fs,lum 5 858 fh,525 5 864 fh,625 5 13.5 MHz
i.e., 525 and 625 line systems have 858 and 864 samples/line, respectively, and for
fs,chr 5 fs,lum/2 5 6.75 MHz
ITU-R BT.601 standards for both 525- and 625-line SDTV systems employ
interlaced scan, where the raw data rate is 165.9 Mbps. The parameters of both formats are shown in Table 2.1. Historically, interlaced SDTV was displayed on analog
cathode ray tube (CRT) monitors, which employ interlaced scanning at 50/60 Hz.
Today, flat-panel displays and projectors can display video at 100/120 Hz interlace
or progressive mode, which requires scan-rate conversion and de-interlacing of the
50i/60i ITU-R BT.601 [ITU 11] broadcast signals.
Recognizing that the resolution of SDTV is well behind today’s technology, a new
high-definition TV (HDTV) standard, ITU-R BT.709-5 [ITU 02], which doubles
the resolution of SDTV in both horizontal and vertical directions, has been approved
with three picture formats: 720p, 1080i, and 1080p. Table 2.1 shows their parameters. Today broadcasters use either 720p/50/60 (called HD) or 1080i/25/29.97
(called FullHD). There are no broadcasts in 1080p format at this time. Note that
many 1080i/25 broadcasts use horizontal sub-sampling to 1440 pixels/line to save
bitrate. 720p/50 format has full temporal resolution 50 progressive frames per
second (with 720 lines). Note that most international HDTV events are captured in
either 1080i/25 or 1080i/29.97 (for 60 Hz countries) and presenting 1080i/29.97
Chapter 2. Digital Images and Video
Table 2.1 ITU-R TV Broadcast Standards
Picture Rate
Aspect Ratio
BT.601-7 480i
2:1 Interlace, 30 Hz (60 fields/s)
4:3, 16:9
BT.601-7 576i
2:1 Interlace, 25 Hz (50 fields/s)
4:3, 16:9
BT.709-5 720p
Progressive, 50 Hz, 60 Hz
BT.709-5 1080i
2:1 Interlace, 25 Hz, 30 Hz
BT.709-5 1080p
BT.2020 2160p
BT.2020 4320p
in 50 Hz countries or vice versa requires scan rate conversion. For 1080i/25 content,
720p/50 broadcasters will need to de-interlace the signal before transmission, and
for 1080i/29.97 content, both de-interlacing and frame-rate conversion is required.
Furthermore, newer 1920 3 1080 progressive scan consumer displays require upscaling 1280 3 720 pixel HD broadcast and 1440 3 1080i/25 sub-sampled FullHD
In the computer and consumer electronics industry, standards for video-display
resolutions are set by a consortia of organizations such as Video Electronics Standards
Association (VESA) and Consumer Electronics Association (CEA). The display standards can be grouped as Video Graphics Array (VGA) and its variants and Extended
Graphics Array (XGA) and its variants. The favorite aspect ratio of the display industry has shifted from the earlier 4:3 to 16:10 and 16:9. Some of these standards are
shown in Table 2.2. The refresh rate was an important parameter for CRT monitors.
Since activated LCD pixels do not flash on/off between frames, LCD monitors do
not exhibit refresh-induced flicker. The only part of an LCD monitor that can produce CRT-like flicker is its backlight, which typically operates at 200 Hz.
Recently, standardization across TV, consumer electronics, and computer industries has started, resulting in the so-called convergence enabled by digital video. For
example, some laptops and cellular phones now feature 1920 3 1080 progressive
mode, which is a format jointly supported by TV, consumer electronics, and computer industries.
Ultra-high definition television (UHDTV) is the most recent standard proposed
by NHK Japan and approved as ITU-R BT.2020 [ITU 12]. It supports the 4K
(2160p) and 8K (4320p) digital-video formats shown in Table 2.1. The Consumer
Electronics Association announced that “ultra high-definition” or “ultra HD” or
2.3 Digital Video
Table 2.2 Display Standards
Aspect Ratio
“UHD” would be used for displays that have an aspect ratio of at least 16:9 and at
least one digital input capable of carrying and presenting native video at a minimum
resolution of 3,840 3 2,160 pixels. The ultra-HD format is very similar to 4K digital
cinema format (see Section 2.5.2) and may become an across industries standard in
the near future.
Video-Interface Standards
Digital-video interface standards enable exchange of uncompressed video between
various consumer electronics devices, including digital TV monitors, computer
monitors, blu-ray devices, and video projectors over cable. Two such standards are
Digital Visual Interface (DVI) and High-Definition Multimedia Interface (HDMI).
HDMI is the most popular interface that enables transfer of video and audio on
a single cable. It is backward compatible with DVI-D or DVI-I. HDMI 1.4 and
higher support 2160p digital cinema and 3D stereo transfer.
Image- and Video-Compression Standards
Various digital-video applications, e.g., SDTV, HDTV, 3DTV, video on demand,
interactive games, and videoconferencing, reach potential users over either broadcast
channels or the Internet. Digital cinema content must be transmitted to movie theatres over satellite links or must be shipped in harddisks. Raw (uncompressed) data
rates for digital video are prohibitive, since uncompressed broadcast HDTV requires
Chapter 2. Digital Images and Video
over 700 Mbits/s and 2K digital cinema data exceeds 5 Gbits/sec in uncompressed
form. Hence, digital video must be stored and transmitted in compressed form,
which leads to compression standards.
Video compression is a key enabling technology for digital video. Standardization
of image and video compression is required to ensure compatibility of digital-video
products and hardware by different vendors. As a result, several video-compression
standards have been developed, and work for even more efficient compression is
ongoing. Major standards for image and video compression are listed in Table 2.3.
Historically, standardization in digital-image communication started with the
ITU-T (formerly CCITT) digital fax standards. The ITU-T Recommendation T.4
using 1D coding for digital fax transmission was ratified in 1980. Later, a more
efficient 2D compression technique was added as an option to the ITU-T recommendation T.30 and ISO JBIG was developed to fix some of the problems with the
ITU-T Group 3 and 4 codes, mainly in the transmission of half-tone images.
JPEG was the first color still-image compression standard. It has also found some
use in frame-by-frame video compression, called motion JPEG, mostly because of
its wide availability in hardware. Later JPEG2000 was developed as a more efficient
alternative especially at low bit rates. However, it has mainly found use in the digital
cinema standards.
The first commercially successful video-compression standard was MPEG-1 for
video storage on CD, which is now obsolete. MPEG-2 was developed for compression of SDTV and HDTV as well as video storage in DVD and was the enabling
technology of digital TV. MPEG-4 AVC and HEVC were later developed as more
efficient compression standards especially for HDTV and UHDTV as well as video
on blu-ray discs. We discuss image- and video-compression technologies and standards in detail in Chapter 7 and Chapter 8, respectively.
Table 2.3 International Standards for Image/Video Compression
ITU-T (formerly CCITT) G3/G4
FAX, Binary images
Binary/halftone, gray-scale images
Still images
Digital cinema
Digital video, SDTV, HDTV
Digital video
2.4 3D Video
2.4 3D Video
3D cinema has gained wide acceptance in theatres as many movies are now produced
in 3D. Flat-panel 3DTV has also been positively received by consumers for watching
sports broadcasts and blu-ray movies. Current 3D-video displays are stereoscopic
and are viewed by special glasses. Stereo-video formats can be classified as framecompatible (mainly for broadcast TV) and full-resolution (sequential) formats.
Alternatively, multi-view and super multi-view 3D-video displays are currently being
developed for autostereoscopic viewing. Multi-view video formats without accompanying depth information require extremely high data rates. Multi-view-plus-depth
representation and compression are often preferred for efficient storage and transmission of multi-view video as the number of views increases. There are also volumetric, holoscopic (integral imaging), and holographic 3D-video formats, which are
mostly considered as futuristic at this time.
The main technical obstacles for 3DTV and video to achieve much wider acceptance at home are: i) developing affordable, free-viewing natural 3D display technologies with high spatial, angular, and depth resolution, and ii) capturing and
producing 3D content in a format that is suitable for these display technologies. We
discuss 3D display technologies and 3D-video formats in more detail below.
2.4.1 3D-Display Technologies
A 3D display should ideally reproduce a light field that is an indistinguishable copy
of the actual 3D scene. However, this is a rather difficult task to achieve with today’s
technology due to very large amounts of data that needs to be captured, processed,
and stored/transmitted. Hence, current 3D displays can only reproduce a limited set
of 3D visual cues instead of the entire light field; namely, they reproduce:
Binocular depth – Binocular disparity in a stereo pair provides relative depth
cue. 3D displays that present only two views, such as stereo TV and digital
cinema, can only provide binocular depth cue.
Head-motion parallax – Viewers expect to see a scene or objects from a slightly
different perspective when they move their head. Multi-view, light-field, or volumetric displays can provide head-motion parallax, although most displays can
provide only limited parallax, such as only horizontal parallax.
We can broadly classify 3D display technologies as multiple-image (stereoscopic and auto-stereoscopic), light-field, and volumetric displays, as summarized in
Chapter 2. Digital Images and Video
(with glasses)
(no glasses)
(no glasses)
(no glasses)
Super multi‐view
Static volume
Holoscopic (Integral)
Swept volume
With head‐tracking
Figure 2.10 Classification of 3D-display technologies.
Figure 2.10. Multiple-image displays present two or more images of a scene by some
multiplexing of color sub-pixels on a planar screen such that the right and left eyes
see two separate images with binocular disparity, and rely upon the brain to fuse the
two images to create the sensation of 3D. Light-field displays present light rays as if
they are originating from a real 3D object/scene using various technologies such that
each pixel of the display can emit multiple light rays with different color, intensity,
and directions, as opposed to multiplexing pixels among different views. Volumetric
displays aim to reconstruct a visual representation of an object/scene using voxels
with three physical dimensions via emission, scattering, or relaying of light from a
well-defined region in the physical (x1, x2, x3) space, as opposed to displaying light
rays emitted from a planar screen.
Multiple-Image Displays
Multiple-image displays can be classified as those that require glasses (stereoscopic)
and those that don’t (auto-stereoscopic).
Stereoscopic displays present two views with binocular disparity, one for the left
and one for the right eye, from a single viewpoint. Glasses are required to ensure that
only the right eye sees the right view and the left eye sees the left view. The glasses
can be passive or active. Passive glasses are used for color (wavelength) or polarization
multiplexing of the two views. Anaglyph is the oldest form of 3D display by color
multiplexing using red and cyan filters. Polarization multiplexing applies horizontal
and vertical (linear), or clockwise and counterclockwise (circular) polarization to the
left and right views, respectively. Glasses apply matching polarization to the right
and left eyes. The display shows both left and right views laid over each other with
polarization matching that of the glasses in every frame. This will lead to some loss of
spatial resolution since half of the sub-pixels in the display panel will be allocated to
the left and right views, respectively, using polarized filters. Active glasses (also called
active shutter) present the left image to only the left eye by blocking the view of the
right eye while the left image is being displayed and vice versa. The display alternates
2.4 3D Video
full-resolution left and right images in sequential order. The active 3D system must
assure proper synchronism between the display and glasses. 3D viewing with passive
or active glasses is the most developed and commercially available form of 3D display
technology. We note that two-view displays lack head-motion parallax and can only
provide 3D viewing from a single point of view (from the point where the right and
left views have actually been captured) no matter from which angle the viewer looks
at the screen. Furthermore, polarization may cause loss of some light due to polarization filter absorption, which may affect scene brightness.
Auto-stereoscopic displays do not require glasses. They can display two views or
multiple views. Separation of views can be achieved by different optics technologies,
such as parallax barriers or lenticular sheets, so that only certain rays are emitted in
certain directions. They can provide head-motion parallax, in addition to binocular
depth cues, by either using head-tracking to display two views generated according
to head/eye position of the viewer or displaying multiple fixed views. In the former,
the need for head-tracking, real-time view generation, and dynamic optics to steer
two views in the direction of the viewer gaze increases hardware complexity. In the
latter, continuous-motion parallax is not possible with a limited number of views,
and proper 3D vision is only possible from some select viewing positions, called
sweet spots. In order to determine the number of views, we divide the head-motion
range into 2 cm intervals (zones) and present a view for each zone. Then, images seen
by the left and right eyes (separated by 6 cm) will be separated by three views. If we
allow 4-5 cm head movement toward the left and right, then the viewing range can
be covered by a total of eight or nine views. The major drawbacks of autostereoscopic
multi-view displays are: i) multiple views are displayed over the same physical screen,
sharing sub-pixels between views in a predetermined pattern, which results in loss of
spatial resolution; ii) cross-talk between multiple views is unavoidable due to limitations of optics; and iii) there may be noticeable parallax jumps from view to view
with a limited number of viewing zones. Due to these reasons, auto-stereoscopic
displays have not entered the mass consumer market yet.
State-of-the art stereoscopic and auto-stereoscopic displays have been reviewed
in [Ure 11]. Detailed analysis of stereoscopic and auto-stereoscopic displays from a
signal-processing perspective and their quality profiles are provided in [Boe 13].
Light-Field and Holographic Displays
Super multi-view (SMV) displays can display up to hundreds of views of a scene
taken from different angles (instead of just a right and left view) to create a seearound effect as the viewer slightly changes his/her viewing (gaze) angle. SMV
displays employ more advanced optical technologies than just allocating certain
Chapter 2. Digital Images and Video
sub-pixels to certain views [Ure 11]. The characteristic parameters of a light-field
display are spatial, angular, and perceived depth resolution. If the number of views
is sufficiently large such that viewing zones are less than 3 mm, two or more views
can be displayed within each eye pupil to overcome the accommodation-vergence
conflict and offer a real 3D viewing experience. Quality measures for 3D light-field
displays have been studied in [Kov 14].
Holographic imaging requires capturing amplitude (intensity), phase differences
(interference pattern), and wavelength (color) of a light field using a coherent light
source (laser). Holoscopic imaging (or integral imaging) does not require a coherent
light source, but employs an array of microlenses to capture and reproduce a 4D
light field, where each lens shows a different view depending on the viewing angle.
Volumetric Displays
Different volumetric display technologies aim at creating a 3D viewing experience
by means of rendering illumination within a volume that is visible to the unaided
eye either directly from the source or via an intermediate surface such as a mirror or glass, which can undergo motion such as oscillation or rotation. They can
be broadly classified as swept-volume displays and static volume displays. Sweptvolume 3D displays rely on the persistence of human vision to fuse a series of slices
of a 3D object, which can be rectangular, disc-shaped, or helical cross-sectioned, into
a single 3D image. Static-volume 3D displays partition a finite volume into addressable volume elements, called voxels, made out of active elements that are transparent
in “off” state but are either opaque or luminous in “on” state. The resolution of a
volumetric display is determined by the number of voxels. It is possible to display
scenes with viewing-position-dependent effects (e.g., occlusion) by including transparency (alpha) values for voxels. However, in this case, the scene may look distorted
if viewed from positions other than those it was generated for.
The light-field, volumetric, and holographic display technologies are still being
developed in major research laboratories around the world and cannot be considered
as mature technologies at the time of writing. Note that light-field and volumetricvideo representations require orders of magnitude more data (and transmission
bandwidth) compared to stereoscopic video. In the following, we cover representations for two-view, multi-view, and super multi-view video.
2.4.2 Stereoscopic Video
Stereoscopic two-view video formats can be classified as frame-compatible and fullresolution formats.
2.4 3D Video
Figure 2.11 Frame compatible formats: (a) side-by-side; (b) top-bottom.
Frame-compatible stereo-video formats have been developed to provide 3DTV
services over existing digital TV broadcast infrastructures. They employ pixel subsampling in order to keep the frame size and rate the same as that of monocular 2D
video. Common sub-sampling patterns include side-by-side, top-and-bottom, line
interleaved, and checkerboard. Side-by-side format, shown in Figure 2.11(a), applies
horizontal subsampling to the left and right views, reducing horizontal resolution
by 50%. The subsampled frames are then put together side-by-side. Likewise, topand-bottom format, shown in Figure 2.11(b), vertically subsamples the left and right
views, and stitches them over-under. In the line-interleaved format, the left and right
views are again sub-sampled vertically, but put together in an interleaved fashion.
Checkerboard format sub-samples left and right views in an offset grid pattern and
multiplexes them into a single frame in a checkerboard layout. Among these formats,
side-by-side and top-and-bottom are selected as mandatory for broadcast by the latest HDMI specification 1.4a [HDM 13]. Frame-compatible formats are also supported by the stereo and multi-view extensions of the most recent joint MPEG and
ITU video-compression standards such as AVC and HEVC (see Chapter 8).
The two-view full resolution stereo is the format of choice for movie and game
content. Frame packing, which is a supported format in the HDMI specification
version 1.4a, stores frames of left and right views sequentially, without any change
in resolution. This full HD stereo-video format requires, in the worst case, twice
as much bandwidth as that of monocular video. The extra bandwidth requirement
may be kept around 50% by using the Multi-View Video Coding (MVC) standard,
which is selected by the Blu-ray Disc Association as the coding format for 3D video.
2.4.3 Multi-View Video
Multi-view and super multi-view displays employ multi-view video representations with varying number of views. Since the required data rate increases linearly with the number of views, depth-based representations are more efficient for
multi-view video with more than a few views. Depth-based representations also
Chapter 2. Digital Images and Video
enable: i) generation of desired intermediate views that are not present among the
original views by using depth-image based rendering (DIBR) techniques, and ii) easy
manipulation of depth effects to adjust vergence vs. accommodation conflict for best
viewing comfort.
View-plus-depth has initially been proposed as a stereo-video format, where a
single view and associated depth map are transmitted to render a stereo pair at the
decoder. It is backward compatible with legacy video using a layered bit stream with
an encoded view and encoded depth map as a supplementary layer. MPEG specified a container format for view-plus-depth data, called MPEG-C Part 3 [MPG 07],
which was later extended to multi-view-video-plus-depth (MVD) format [Smo 11],
where N views and N depth maps are encoded and transmitted to generate M views
at the decoder, with N  M. The MVD format is illustrated in Figure 2.12, where
only 6 views and 6 depth maps per frame are encoded to reconstruct 45 views per
frame at the decoder side by using DIBR techniques.
The depth information needs to be accurately captured/computed, encoded, and
transmitted in order to render intermediate views accurately using the received reference view and depth map. Each frame of the depth map conveys the distance of the
corresponding video pixel from the camera. Scaled depth values, represented by 8
bits, can be regarded as a separate gray-scale video, which can be compressed very
efficiently using state-of-the-art video codecs. Depth map typically requires 15–20%
45 Virtual Intermediate Views
Figure 2.12 N-view 1 N depth-map format (courtesy of Aljoscha Smolic).
2.5 Digital-Video Applications
of the bitrate necessary to encode the original video due to its smooth and lessstructured nature.
A difficulty with the view-plus-depth format is generation of accurate depth
maps. Although there are time-of-flight cameras that can generate depth or disparity maps, they typically offer limited performance in outdoors environments.
Algorithms for depth and disparity estimation by image rectification and disparity
matching have been studied in the literature [Kau 07]. Another difficulty is the
appearance of regions in the rendered views, which are occluded in the available
views. These disocclusion regions may be concealed by smoothing the original depthmap data to avoid appearance of holes. Also, it is possible to use multiple view-plusdepth data to prevent disocclusions [Mul 11]. An extension of the view-plus-depth,
which allows better modeling of occlusions, is the layered depth video (LDV). LDV
provides multiple depth values for each pixel in a video frame.
While high-definition digital-video products have gained universal user acceptance, there are a number of challenges to overcome in bringing 3D video to consumers. Most importantly, advances in autostereoscopic (without glasses) multi-view
display technology will be critical for practical usability and consumer acceptance of
3D viewing technology. Availability of high-quality 3D content at home is another
critical factor. In summary, both content creators and display manufacturers need
further effort to provide consumers with a high-quality 3D experience without viewing discomfort or fatigue and high transition costs. It seems that the TV/consumer
electronics industry has moved its focus to bringing ultra-high-definition products
to consumers until there is more progress with these challenges.
2.5 Digital-Video Applications
Main consumer applications for digital video include digital TV broadcasts, digital
cinema, video playback from DVD or blu-ray players, as well as video streaming and
videoconferencing over the Internet (wired or wireless) [Pit 13].
2.5.1 Digital TV
A digital TV (DTV) broadcasting system consists of video/audio compression, multiplex and transport protocols, channel coding, and modulation subsystems. The
biggest single innovation that enabled digital TV services has been advances in video
compression since the 1990s. Video-compression standards and algorithms are covered in detail in Chapter 8. Video and audio are compressed separately by different
encoders to produce video and audio packetized elementary streams (PES). Video and
Chapter 2. Digital Images and Video
audio PES and related data are multiplexed into an MPEG program stream (PS).
Next, one or more PSs are multiplexed into an MPEG transport stream (TS). TS
packets are 188-bytes long and are designed with synchronization and recovery in
mind for transmission in lossy environments. The TS is then modulated into a signal
for transmission. Several different modulation methods exist that are specific to the
medium of transmission, which are terrestial (fixed reception), cable, satellite, and
mobile reception.
There are different digital TV broadcasting standards that are deployed globally.
Although they all use MPEG-2 or MPEG-4 AVC/H.264 video compression, more
or less similar audio coding, and the same transport stream protocol, their channel coding, transmission bandwidth and modulation systems differ slightly. These
include the Advanced Television System Committee (ATSC) in the USA, Digital
Video Broadcasting (DVB) in Europe, Integrated Multimedia Broadcasting (ISDB)
in Japan, and Digital Terrestial Multimedia Broadcasting in China.
ATSC Standards
The first DTV standard was ATSC Standard A/53, which was published in 1995
and was adopted by the Federal Communications Commission in the United
States in 1996. This standard supported MPEG-2 Main profile video encoding
and 5.1-channel surround sound using Dolby Digital AC-3 encoding, which was
standardized as A/52. Support for AVC/H.264 video encoding was added with the
ATSC Standard A/72 that was approved in 2008. ATSC signals are designed to use
the same 6 MHz bandwidth analog NTSC television channels. Once the digital
video and audio signals have been compressed and multiplexed, ATSC uses a 188byte MPEG transport stream to encapsulate and carry several video and audio programs and metadata. The transport stream is modulated differently depending on
the method of transmission:
Terrestrial broadcasters use 8-VSB modulation that can transmit at a maximum
rate of 19.39 Mbit/s. ATSC 8-VSB transmission system adds 20 bytes of ReedSolomon forward-error correction to create packets that are 208 bytes long.
Cable television stations operate at a higher signal-to-noise ratio than terrestial broadcasters and can use either 16-VSB (defined by ATSC) or 256-QAM
(defined by Society of Cable Telecommunication Engineers) modulation to
achieve a throughput of 38.78 Mbit/s, using the same 6-MHz channel.
There is also an ATSC standard for satellite transmission; however, directbroadcast satellite systems in the United States and Canada have long used
2.5 Digital-Video Applications
either DVB-S (in standard or modified form) or a proprietary system such as
DSS (Hughes) or DigiCipher 2 (Motorola).
The receiver must demodulate and apply error correction to the signal. Then, the
transport stream may be de-multiplexed into its constituent streams before audio
and video decoding.
The newest edition of the standard is ATSC-3.0, which employs the HEVC/H.265
video codec, with OFDM instead of 8-VSB for terrestial modulation, allowing for
28 Mbps or more of bandwidth on a single 6-MHz channel.
DVB Standards
DVB is a suite of standards, adopted by the European Telecommunications Standards Institute (ETSI) and supported by European Broadcasting Union (EBU),
which defines the physical layer and data-link layer of the distribution system. The
DVB texts are available on the ETSI website. They are specific for each medium of
transmission, which we briefly review.
DVB-T and DVB-T2
DVB-T is the DVB standard for terrestrial broadcast of digital television and was first
published in 1997. It specifies transmission of MPEG transport streams, containing
MPEG-2 or H.264/MPEG-4 AVC compressed video, MPEG-2 or Dolby Digital
AC-3 audio, and related data, using coded orthogonal frequency-division multiplexing (COFDM) or OFDM modulation. Rather than carrying data on a single radio
frequency (RF) channel, COFDM splits the digital data stream into a large number
of lower rate streams, each of which digitally modulates a set of closely spaced adjacent
sub-carrier frequencies. There are two modes: 2K-mode (1,705 sub-carriers that are
4 kHz apart) and 8K-mode (6,817 sub-carriers that are 1 kHz apart). DVB-T offers
three different modulation schemes (QPSK, 16QAM, 64QAM). It was intended for
DTV broadcasting using mainly VHF 7 MHz and UHF 8 MHz channels. The first
DVB-T broadcast was realized in the UK in 1998. The DVB-T2 is the extension
of DVB-T that was published in June 2008. With several technical improvements,
it provides a minimum 30% increase in payload, under similar channel conditions
compared to DVB-T. The ETSI adopted the DVB-T2 in September 2009.
DVB-S and DVB-S2
DVB-S is the original DVB standard for satellite television. Its first release dates back
to 1995, while development lasted until 1997. The standard only specifies physical
Chapter 2. Digital Images and Video
link characteristics and framing for delivery of MPEG transport stream (MPEG-TS)
containing MPEG-2 compressed video, MPEG-2 or Dolby Digital AC-3 audio,
and related data. The first commercial application was in Australia, enabling digitally broadcast, satellite-delivered television to the public. DVB-S has been used in
both multiple-channel per carrier and single-channel per carrier modes for broadcast
network feeds and direct broadcast satellite services in every continent of the world,
including Europe, the United States, and Canada.
DVB-S2 is the successor of the DVB-S standard. It was developed in 2003 and
ratified by the ETSI in March 2005. DVB-S2 supports broadcast services including
standard and HDTV, interactive services including Internet access, and professional
data content distribution. The development of DVB-S2 coincided with the introduction of HDTV and H.264 (MPEG-4 AVC) video codecs. Two new key features
that were added compared to the DVB-S standard are:
A powerful coding scheme, Irregular Repeat-Accumulate codes, based on a
modern LDPC code, with a special structure for low encoding complexity.
Variable coding and modulation (VCM) and adaptive coding and modulation (ACM) modes to optimize bandwidth utilization by dynamically changing
transmission parameters.
Other features include enhanced modulation schemes up to 32-APSK, additional code rates, and introduction of a generic transport mechanism for IP packet
data including MPEG-4 AVC video and audio streams, while supporting backward
compatibility with existing DVB-S transmission. The measured DVB-S2 performance gain over DVB-S is around a 30% increase of available bitrate at the same
satellite transponder bandwidth and emitted signal power. With improvements in
video compression, an MPEG-4 AVC HDTV service can now be delivered in the
same bandwidth used for an early DVB-S based MPEG-2 SDTV service. In March
2014, the DVB-S2X specification was published as an optional extension adding
further improvements.
DVB-C and DVB-C2
The DVB-C standard is for broadcast transmission of digital television over cable.
This system transmits an MPEG-2 or MPEG-4 family of digital audio/digital video
stream using QAM modulation with channel coding. The standard was first published by the ETSI in 1994, and became the most widely used transmission system
for digital cable television in Europe. It is deployed worldwide in systems ranging
2.5 Digital-Video Applications
from larger cable television networks (CATV) to smaller satellite master antenna TV
(SMATV) systems.
The second-generation DVB cable transmission system DVB-C2 specification
was approved in April 2009. DVB-C2 allows bitrates up to 83.1 Mbit/s on an 8
MHz channel when using 4096-QAM modulation, and up to 97 Mbit/s and 110.8
Mbit/s per channel when using 16384-QAM and 65536-AQAM modulation,
respectively. By using state-of-the-art coding and modulation techniques, DVB-C2
offers more than a 30% higher spectrum efficiency under the same conditions, and
the gains in downstream channel capacity are greater than 60% for optimized HFC
networks. These results show that the performance of the DVB-C2 system gets so
close to the theoretical Shannon limit that any further improvements would most
likely not be able to justify the introduction of a disruptive third generation cabletransmission system.
There is also a DVB-H standard for terrestrial mobile TV broadcasting to handheld devices. The competitors of this technology have been the 3G cellular-systembased MBMS mobile-TV standard, the ATSC-M/H format in the United States,
and the Qualcomm MediaFLO. DVB-SH (satellite to handhelds) and DVB-NGH
(Next Generation Handheld) are possible future enhancements to DVB-H. However, none of these technologies have been commercially successful.
2.5.2 Digital Cinema
Digital cinema refers to digital distribution and projection of motion pictures as
opposed to use of motion picture film. A digital cinema theatre requires a digital projector (instead of a conventional film projector) and a special computer server. Movies are supplied to theatres as digital files, called a Digital Cinema Package (DCP),
whose size is between 90 gigabytes (GB) and 300 GB for a typical feature movie. The
DCP may be physically delivered on a hard drive or can be downloaded via satellite.
The encrypted DCP file first needs to be copied onto the server. The decryption keys,
which expire at the end of the agreed upon screening period, are supplied separately
by the distributor. The keys are locked to the server and projector that will screen the
film; hence, a new set of keys are required to show the movie on another screen. The
playback of the content is controlled by the server using a playlist.
Technology and Standards
Digital cinema projection was first demonstrated in the United States in October
1998 using Texas Instruments’ DLP projection technology. In January 2000, the
Chapter 2. Digital Images and Video
Society of Motion Picture and Television Engineers, in North America, initiated a
group to develop digital cinema standards. The Digital Cinema Initiative (DCI), a
joint venture of six major studios, was established in March 2002 to develop a system
specification for digital cinema to provide robust intellectual property protection
for content providers. DCI published the first version of a specification for digital
cinema in July 2005. Any DCI-compliant content can play on any DCI-compliant
hardware anywhere in the world.
Digital cinema uses high-definition video standards, aspect ratios, or frame rates
that are slightly different than HDTV and UHDTV. The DCI specification supports 2K (2048 3 1080 or 2.2 Mpixels) at 24 or 48 frames/sec and 4K (4096 3 2160
or 8.8 Mpixels) at 24 frames/sec modes, where resolutions are represented by the
horizontal pixel count. The 48 frames/sec is called high frame rate (HFR). The specification employs the ISO/IEC 15444-1 JPEG2000 standard for picture encoding,
and the CIE XYZ color space is used at 12 bits per component encoded with a 2.6
gamma applied at projection. It ensures that 2K content can play on 4K projectors
and vice versa.
Digital Cinema Projectors
Digital cinema projectors are similar in principle to other digital projectors used in
the industry. However, they must be approved by the DCI for compliance with the
DCI specifications: i) they must conform to the strict performance requirements,
and ii) they must incorporate anti-piracy protection to protect copyrights. Major
DCI-approved digital cinema projector manufacturers include Christie, Barco,
NEC, and Sony. The first three manufactuers have licensed the DLP technology
from Texas Instruments, and Sony uses its own SXRD technology. DLP projectors
were initially available in 2K mode only. DLP projectors became available in both
2K and 4K in early 2012, when Texas Instruments’ 4K DLP chip was launched.
Sony SXRD projectors are only manufactured in 4K mode.
DLP technology is based on digital micromirror devices (DMDs), which are
chips whose surface is covered by a large number of microscopic mirrors, one for
each pixel; hence, a 2K chip has about 2.2 million mirrors and a 4K chip about 8.8
million. Each mirror vibrates several thousand times a second between on and off
positions. The proportion of the time the mirror is in each position varies according
to the brightness of each pixel. Three DMD devices are used for color projection,
one for each of the primary colors. Light from a Xenon lamp, with power between
1 kW and 7 kW, is split by color filters into red, green, and blue beams that are
directed at the appropriate DMD.
2.5 Digital-Video Applications
Transition to digital projection in cinemas is ongoing worldwide. According to
the National Association of Theatre Owners, 37,711 screens out of 40,048 in the
United States had been converted to digital and about 15,000 were 3D capable as
of May 2014.
3D Digital Cinema
The number of 3D-capable digital cinema theatres is increasing with wide interest of
audiences in 3D movies and an increasing number of 3D productions. A 3D-capable
digital cinema video projector projects right-eye and left-eye frames sequentially. The
source video is produced at 24 frames/sec per eye; hence, a total of 48 frames/sec
for right and left eyes. Each frame is projected three times to reduce flicker, called
triple flash, for a total of 144 times per second. A silver screen is used to maintain
light polarization upon reflection. There are two types of stereoscopic 3D viewing
technology where each eye sees only its designated frame: i) glasses with polarizing
filters oriented to match projector filters, and ii) glasses with liquid crystal (LCD)
shutters that block or transmit light in sync with the projectors. These technologies
are provided under the brands RealD, MasterImage, Dolby 3D, and XpanD.
The polarization technology combines a single 144-Hz digital projector with
either a polarizing filter (for use with polarized glasses and silver screens) or a filter
wheel. RealD 3D cinema technology places a push-pull electro-optical liquid crystal
modulator called a ZScreen in front of the projector lens to alternately polarize each
frame. It circularly polarizes frames clockwise for the right eye and counter-clockwise
for the left eye. MasterImage uses a filter wheel that changes the polarity of the projector’s light output several times per second to alternate the left-and-right-eye views.
Dolby 3D also uses a filter wheel. The wheel changes the wavelengths of colors being
displayed, and tinted glasses filter these changes so the incorrect wavelength cannot
enter the wrong eye. The advantage of circular polarization over linear polarization
is that viewers are able to slightly tilt their head without seeing double or darkened
The XpanD system alternately flashes the images for each eye that viewers observe
using electronically synchronized glasses The viewer wears electronic glasses whose
LCD lenses alternate between clear and opaque to show only the correct image at the
correct time for each eye. XpanD uses an external emitter that broadcasts an invisible infrared signal in the auditorium that is picked up by glasses to synchronize the
shutter effect.
IMAX Digital 3D uses two separate 2K projectors that represent the left and right
eyes. They are separated by a distance of 64 mm (2.5 in), which is the average distance
Chapter 2. Digital Images and Video
between a human’s eyes. The two 2K images are projected over each other (superposed) on a silver screen with proper polarization, which makes the image brighter.
Right and left frames on the screen are directed only to the correct eye by means of
polarized glasses that enable the viewer to see in 3D. Note that IMAX theatres use
the original 15/70 IMAX higher resolution frame format on larger screens.
2.5.3 Video Streaming over the Internet
Video streaming refers to delivery of media over the Internet, where the client player
can begin playback before the entire file has been sent by the server. A server-client
streaming system consists of a streaming server and a client that communicate using
a set of standard protocols. The client may be a standalone player or a plugin as
part of a Web browser. The streaming session can be a video-on-demand request
(sometimes called a pull-application) or live Internet broadcasting (called a pushapplication). In a video-on-demand session, the server streams from a pre-encoded
and stored file. Live streaming refers to live content delivered in real-time over the
Internet, which requires a live camera and a real-time encoder on the server side.
Since the Internet is a best-effort channel, packets may be delayed or dropped by
the routers and the effective end-to-end bitrates fluctuate in time. Adaptive streaming technologies aim to adapt the video-source (encoding) rate according to an estimate of the available end-to-end network rate. One possible way to do this is stream
switching, where the server encodes source video at multiple pre-selected bitrates and
the client requests switching to the stream encoded at the rate that is closest to its
network access rate. A less commonly deployed solution is based on scalable video
coding, where one or more enhancement layers of video may be dropped to reduce
the bitrate as needed.
In the server-client model, the server sends a different stream to each client.
This model is not scalable, since server load increases linearly with the number of
stream requests. Two solutions to solve this problem are multicasting and peer-topeer (P2P) streaming. We discuss the server-client, multicast, and P2P streaming
models in more detail below.
Server-Client Streaming
This is the most commonly used streaming model on the Internet today. All video
streaming systems deliver video and audio streams by using a streaming protocol
built on top of transmission control protocol (TCP) or user datagram protocol
(UDP). Streaming solutions may be based on open-standard protocols published by
2.5 Digital-Video Applications
the Internet Engineering Task Force (IETF) such as RTP/UDP or HTTP/TCP, or
may be proprietary systems, where RTP stands for real-time transport protocol and
HTTP stands for hyper-text transfer protocol.
Streaming Protocols
Two popular streaming protocols are Real-Time Streaming Protocol (RTSP), an
open standard developed and published by the IETF as RFC 2326 in 1998, and
Real Time Messaging Protocol (RTMP), a proprietary solution developed by Adobe
RTSP servers use the Real-time Transport Protocol (RTP) for media stream
delivery, which supports a range of media formats (such as AVC/H.264, MJPEG,
etc.). Client applications include QuickTime, Skype, and Windows Media Player.
Android smartphone platforms also include support for RTSP as part of the 3GPP
RTMP is primarily used to stream audio and video to Adobe’s Flash Player client.
The majority of streaming videos on the Internet is currently delivered via RTMP
or one of its variants due to the success of the Flash Player. RTMP has been released
for public use. Adobe has included support for adaptive streaming into the RTMP
The main problem with UDP-based streaming is that streams are frequently
blocked by firewalls, since they are not being sent over HTTP (port 80). In order
to circumvent this problem, protocols have been extended to allow for a stream to
be encapsulated within HTTP requests, which is called tunneling. However, tunneling comes at a performance cost and is often only deployed as a fallback solution. Streaming protocols also have secure variants that use encryption to protect
the stream.
HTTP Streaming
Streaming over HTTP, which is a more recent technology, works by breaking a
stream into a sequence of small HTTP-based file downloads, where each download loads one short chunk of the whole stream. All flavors of HTTP streaming
include support for adaptive streaming (bitrate switching), which allows clients to
dynamically switch between different streams of varying quality and chunk size during playback, in order to adapt to changing network conditions and available CPU
resources. By using HTTP, firewall issues are generally avoided. Another advantage
of HTTP streaming is that it allows HTTP chunks to be cached within ISPs or
Chapter 2. Digital Images and Video
corporations, which would reduce the bandwidth required to deliver HTTP streams,
in contrast to video streamed via RTMP.
Different vendors have implemented different HTTP-based streaming solutions,
which all use similar mechanisms but are incompatible; hence, they all require the
vendor’s own software:
HTTP Live Streaming (HLS) by Apple is an HTTP-based media streaming
protocol that can dynamically adjust movie playback quality to match the available speed of wired or wireless networks. HTTP Live Streaming can deliver
streaming media to an iOS app or HTML5-based website. It is available as an
IETF Draft (as of October 2014) [Pan 14].
Smooth Streaming by Microsoft enables adaptive streaming of media to clients
over HTTP. The format specification is based on the ISO base media file format. Microsoft provides Smooth Streaming Client software development kits
for Silverlight and Windows Phone 7.
HTTP Dynamic Streaming (HDS) by Adobe provides HTTP-based adaptive
streaming of high-quality AVC/H.264 or VP6 video for a Flash Player client
MPEG-DASH is the first adaptive bit-rate HTTP-based streaming solution
that is an international standard, published in April 2012. MPEG-DASH is audio/
video codec agnostic. It allows devices such as Internet-connected televisions, TV
set-top boxes, desktop computers, smartphones, tablets, etc., to consume multimedia delivered via the Internet using previously existing HTTP web server
infrastructure, with the help of adaptive streaming technology. Standardizing an
adaptive streaming solution aims to provide confidence that the solution can be
adopted for universal deployment, compared to similar proprietary solutions such
as HLS by Apple, Smooth Streaming by Microsoft, or HDS by Adobe. An implementation of MPEG-DASH using a content centric networking (CCN) naming
scheme to identify content segments is publicly available [Led 13]. Several issues
still need to be resolved, including legal patent claims, before DASH can become
a widely used standard.
Multicast and Peer-to-Peer (P2P) Streaming
Multicast is a one-to-many delivery system, where the source server sends each packet
only once, and the nodes in the network replicate packets only when necessary to
reach multiple clients. The client nodes send join and leave messages, e.g., as in the
2.5 Digital-Video Applications
case of Internet television when the user changes the TV channel. In P2P streaming, clients (peers) forward packets to other peers (as opposed to network nodes) to
minimize the load on the source server.
The multicast concept can be implemented at the IP or application level. The
most common transport layer protocol to use multicast addressing is the User Datagram Protocol (UDP). IP multicast is implemented at the IP routing level, where
routers create optimal distribution paths for datagrams sent to a multicast destination address. IP multicast has been deployed in enterprise networks and multimedia
content delivery networks, e.g., in IPTV applications. However, IP multicast is not
implemented in commercial Internet backbones mainly due to economic reasons.
Instead, application layer multicast-over-unicast overlay services for application-level
group communication are widely used.
In media streaming over P2P overlay networks, each peer forwards packets to
other peers in a live media streaming session to minimize the load on the server.
Several protocols that help peers find a relay peer for a specified stream exist [Gu 14].
There are P2PTV networks based on real-time versions of the popular file-sharing
protocol BitTorrent. Some P2P technologies employ the multicast concept when
distributing content to multiple recipients, which is known as peercasting.
2.5.4 Computer Vision and Scene/Activity Understanding
Computer vision is a discipline of computer science that aims to duplicate abilities
of human vision by processing and understanding digital images and video. It is such
a large field that it is the subject of many excellent textbooks [Har 04, For 11, Sze
11]. The visual data to be processed can be still images, video sequences, or views
from multiple cameras. Computer vision is generally divided into high-level and
low-level vision. High-level vision is often considered as part of artificial intelligence
and is concerned with the theory of learning and pattern recognition with application to object/activity recognition in order to extract information from images and
video. We mention computer vision here because many of the problems addressed in
image/video processing and low-level vision are common. Low-level vision includes
many image- and video-processing tasks that are the subject of this book such as
edge detection, image enhancement and restoration, motion estimation, 3D scene
reconstruction, image segmentation, and video tracking. These low-level vision tasks
have been used in many computer-vision applications, including road monitoring,
military surveillance, and robot navigation. Indeed, several of the methods discussed
in this book have been developed by computer-vision researchers.
Chapter 2. Digital Images and Video
2.6 Image and Video Quality
Video quality may be measured by the quality of experience of viewers, which can
usually be reliably measured by subjective methods. There have been many studies to
develop objective measures of video quality that correlate well with subjective evaluation results [Cho 14, Bov 13]. However, this is still an active research area. Since
analog video is becoming obsolete, we start by defining some visual artifacts related
to digital video that are the main cause of loss of quality of experience.
2.6.1 Visual Artifacts
Artifacts are visible distortions in images/videos. We can classify visual artifacts as
spatial and temporal artifacts. Spatial artifacts, such as blur, noise, ringing, and blocking, are most disturbing in still images but may also be visible in video. In addition,
in video, temporal freeze and skipped frames are important causes of visual disturbance and, hence, loss of quality of experience.
Blur refers to lack or loss of image sharpness (high spatial frequencies). The main
causes of blur are insufficient spatial resolution, defocus, and/or motion between
camera and the subject. According to the Nyquist sampling theorem, the highest
horizontal and vertical spatial frequencies that can be represented is determined by
the sampling rate (pixels/cm), which relates to image resolution. Consequently, lowresolution images cannot contain high spatial frequencies and appear blurred. Defocus blur is due to incorrect focus of the camera, which may be due to depth of field.
Motion blur is caused by relative movement of the subject and camera while the
shutter is open. It may be more noticeable in imaging darker scenes since the shutter
has to remain open for longer time.
Image noise refers to low amplitude, high-frequency random fluctuations in the
pixel values of recorded images. It is an undesirable by-product of image capture,
which can be produced by film grain, photo-electric sensors, and digital camera
circuitry, or image compression. It is measured by signal-to-noise ratio. Noise due
to electronic fluctuations can be modeled by a white, Gaussian random field, while
noise due to LCD sensor imperfections is usually modeled as impulsive (salt-andpepper) noise. Noise at low-light (signal) levels can be modeled as speckle noise.
Image/video compression also generates noise, known as quantization noise and
mosquito noise. Quantization or truncation of the DCT/wavelet transform coefficients results in quantization noise. Mosquito noise is temporal noise, i.e., flickeringlike luminance/chrominance fluctuations as a consequence of differences in coding
observed in smoothly textured regions or around high contrast edges in consecutive
frames of video.
2.6 Image and Video Quality
Ringing and blocking artifacts, which are by-products of DCT image/video
compression, are also observed in compressed images/video. Ringing refers to oscillations around sharp edges. It is caused by sudden truncation of DCT coefficients
due to coarse quantization (also known as the Gibbs effect). DCT is usually taken
over 8 3 8 blocks. Coarse quantization of DC coefficients may cause mismatch of
image mean over 8 3 8 blocks, which results in visible block boundaries known as
blocking artifacts.
Skip frame and freeze frame are the result of video transmission over unreliable
channels. They are caused by video packets that are not delivered on time. When
video packets are late, there are two options: skip late packets and continue with the
next packet, which is delivered on time, or wait (freeze) until the late packets arrive.
Skipped frames result in motion jerkiness and discontinuity, while freeze frame refers
to complete stopping of action until the video is rebuffered.
Visibility of artifacts is affected by the viewing conditions, as well as the type of
image/video content as a result of spatial and temporal-masking effects. For example,
spatial-image artifacts that are not visible in full-motion video may be higly objectionable when we freeze frame.
2.6.2 Subjective Quality Assessment
Measurement of subjective video quality can be challenging because many parameters of set-up and viewing conditions, such as room illumination, display type,
brightness, contrast, resolution, viewing distance, and the age and educational level
of experts, can influence the results. The selection of video content and the duration
also affect the results. A typical subjective video quality evaluation procedure consists
of the following steps:
1. Choose video sequences for testing
2. Choose the test set-up and settings of system to evaluate
3. Choose a test method (how sequences are presented to experts and how their
opinion is collected: DSIS, DSCQS, SSCQE, DSCS)
4. Invite sufficient number and types of experts (18 or more is recommended)
5. Carry out testing and calculate the mean expert opinion scores (MOS) for each
test set-up
In order to establish meaningful subjective assessment results, some test methods,
grading scales, and viewing conditions have been standardized by ITU-T Recommendation BT.500-11 (2002) “Methodology for the subjective assessment of the
quality of television pictures.” Some of these test methods are double stimulus where
Chapter 2. Digital Images and Video
viewers rate the quality or change in quality between two video streams (reference
and impaired). Others are single stimulus where viewers rate the quality of just one
video stream (the impaired). Examples of the former are the double stimulus impairment scale (DSIS), double stimulus continuous quality scale (DSCQS), and double
stimulus comparison scale (DSCS) methods. An example of the latter is the single
stimulus continuous quality evaluation (SSCQE) method. In the DSIS method,
observers are first presented with an unimpaired reference video, then the same video
impaired, and he/she is asked to vote on the second video using an impairment
scale (from “impairments are imperceptible” to “impairments are very annoying”).
In the DSCQS method, the sequences are again presented in pairs: the reference and
impaired. However, observers are not told which one is the reference and are asked to
assess the quality of both. In the series of tests, the position of the reference is changed
randomly. Different test methodologies have claimed advantages for different cases.
2.6.3 Objective Quality Assessment
The goal of objective image quality assessment is to develop quantitative measures that
can automatically predict perceived image quality [Bov 13]. Objective image/video
quality metrics are mathematical models or equations whose results are expected to
correlate well with subjective assessments. The goodness of an objective video-quality
metric can be assessed by computing the correlation between the objective scores and
the subjective test results. The most frequently used correlation coefficients are the
Pearson linear correlation coefficient, Spearman rank-order correlation coefficient,
kurtosis, and the outliers ratio.
Objective metrics are classified as full reference (FR), reduced reference (RR),
and no-reference (NR) metrics, based on availability of the original (high-quality)
video, which is called the reference. FR metrics compute a function of the difference
between every pixel in each frame of the test video and its corresponding pixel in the
reference video. They cannot be used to evaluate the quality of the received video,
since a reference video is not available at the receiver end. RR metrics extract some
features of both videos and compare them to give a quality score. Only some features
of the reference video must be sent along with the compressed video in order to
evaluate the received video quality at the receiver end. NR metrics assess the quality
of a test video without any reference to the original video.
Objective Image/Video Quality Measures
Perhaps the most well-established methodology for FR objective image and video
quality evaluation is pixel-by-pixel comparison of image/video with the reference.
2.6 Image and Video Quality
The peak signal-to-noise ratio (PSNR) measures the logarithm of the ratio of the
maximum signal power to the mean square difference (MSE), given by
 2552 
PSNR 5 10 log10 
 MSE 
where the MSE between the test video sˆ[ n1 , n2 , k ], which is N1 3 N2 pixels and N3
frames long, and reference video s [ n1 , n2 , k ] with the same size, can be computed by
∑∑∑( s[n1 , n2 , k ] sˆ[n1 , n2 , k ])2
N1N 2 N3 n10 n2 0 k0
Some have claimed that PSNR may not correlate well with the perceived visual
quality since it does not take into account many characteristics of the human visual
system, such as spatial- and temporal-masking effects. To this effect, many alternative FR metrics have been proposed. They can be classified as those based on structural similarity and those based on human vision models.
The structural similarity index (SSIM) is a structural image similarity based FR
metric that aims to measure perceived change in structural information between two
N 3 N luminance blocks x and y, with means mx and my and variances s x2 and s 2y ,
respectively. It is given by [Wan 04]
SSIM ( x , y ) (2 x y c1 )(2 xy c 2 )
(x2 2y c1 )( x2 2y c 2 )
where s xy is the covariance between windows x and y and c1 and c2 are small constants to avoid division by very small numbers.
Perceptual evaluation of video quality (PEVQ) is a vision-model-based FR metric that analyzes pictures pixel-by-pixel after a temporal alignment (registration) of
corresponding frames of reference and test video. PEVQ aims to reflect how human
viewers would evaluate video quality based on subjective comparison and outputs
mean opinion scores (MOS) in the range from 1 (bad) to 5 (excellent).
VQM is an RR metric that is based on a general model and associated calibration
techniques and provides estimates of the overall impressions of subjective video quality [Pin 04]. It combines perceptual effects of video artifacts including blur, noise,
blockiness, color distortions, and motion jerkiness into a single metric.
NR metrics can be used for monitoring quality of compressed images/video
or video streaming over the Internet. Specific NR metrics have been developed for
Chapter 2. Digital Images and Video
quantifying such image artifacts as noise, blockiness, and ringing. However, the ability of these metrics to make accurate quality predictions are usually satisfactory only
in a limited scope, such as for JPEG/JPEG2000 images.
The International Telecommunications Union (ITU) Video Quality Experts
Group (VQEG) standardized some of these metrics, including the PEVQ, SSIM,
and VQM, as ITU-T Rec. J.246 (RR) and J.247 (FR) in 2008 and ITU-T Rec.
J.341 (FR HD) in 2011. It is perhaps useful to distinguish the performance of these
structural similarity and human vision model based metrics on still images and
video. It is fair to say these metrics have so far been more successful on still images
than video for objective quality assessment.
Objective Quality Measures for Stereoscopic 3D Video
FR metrics for evaluation of 3D image/video quality is technically not possible, since
the 3D signal is formed only in the brain. Hence, objective measures based on a stereo pair or video-plus-depth-maps should be considered as RR metrics. It is generally
agreed upon that 3D quality of experience is related to at least three factors:
Quality of display technology (cross-talk)
Quality of content (visual discomfort due to accomodation-vergence conflict)
Encoding/transmission distortions/ artifacts
In addition to those artifacts discussed in Section 2.6.1, the main factors in 3D
video quality of experience are visual discomfort and depth perception. As discussed
in Section 2.1.4, visual discomfort is mainly due to the conflict between accommodation and vergence and cross-talk between the left and right views. Human
perception of distortions/artifacts in 3D stereo viewing is not fully understood yet.
There have been some preliminary works on quantifying visual comfort and depth
perception [Uka 08, Sha 13]. An overview of evaluation of stereo and multi-view
image/video quality can be found in [Win 13]. There are also some studies evaluating the perceptual quality of symmetrically and asymmetrically encoded stereoscopic
videos [Sil 13].
[Boe 13] Boev, A., R. Bregovic, and A. Gotchev, “Signal processing for stereoscopic
and multiview 3D displays,” chapter in Handbook of Signal Processing Systems,
Second Edition, (ed. S. Bhattacharyya, E. Deprettere, R. Leupers, and J. Takala),
pp. 3–47, New York, NY: Springer, 2013.
[Bov 13] Bovik, A. C., “Automatic prediction of perceptual image and video quality,”
Proc. of the IEEE, vol. 101, no. 9, pp. 2008–2024, Sept. 2013.
[Bre 07] Breitmeyer, B. G., “Visual masking: past accomplishments, present status,
future developments,” Adv. Cogn. Psychol 3, 2007.
[Cho 14] Choi, L. K., Y. Liao, A. C. Bovik, “Video QoE models for the compute
continuum,” IEEE ComSoc MMTC E-Letter, vol. 8, no. 5, pp. 26–29, Sept. 2013.
[Dub 10] Dubois, E., The Structure and Properties of Color Spaces and the Representation
of Color Images, Morgan & Claypool, 2010.
[Fer 01] Ferwerda, J. A., “Elements of early vision for computer graphics, IEEE
Computer Graphics and Application, vol. 21, no. 5, pp. 22–33, Sept./
Oct. 2001.
[For 11] Forsyth, David A. and Jean Ponce, Computer Vision: A Modern Approach,
Second Edition, Upper Saddle River, NJ: Prentice Hall, 2011.
[Gon 07] Gonzalez, Rafael C. and Richard E. Woods, Digital Image Processing, Third
Edition, Upper Saddle River, NJ: Prentice Hall, 2007.
[Gra 10] Granados, M., B. Ajdin, M. Wand, C. Theobalt, H-P. Seidel, and H. P.A.
Lensch, “Optimal HDR reconstruction with linear digital cameras,” IEEE
Int. Conf. Computer Vision and Pattern Recognition (CVPR), pp. 215–222,
June 2010.
[Gu 14] Gu, Y., et al., Survey of P2P Streaming Applications, IETF draft-ietf-ppspsurvey-08, April 2014.
[Har 04] Hartley, R. I. and A. Zisserman, Multiple View Geometry in Computer
Vision, Second Edition, New York, NY: Cambridge University Press, 2004.
[HDM 13] High definition multimedia interface (HDMI).
[Hub 88] Hubel, D. H., “Eye, Brain, and Vision, Vol. 22, Scientific American
Library,” distributed by W. H. Freeman & Co., New York, 1988. http://hubel
[IEC 00] IEC 61966-2-1:2000, Multimedia systems and equipment - Colour
measurement and management - Colour management - Default RGB colour
space: sRGB, Sept. 2000.
[ITU 02] ITU-R Rec. BT.709, Parameter values for the HDTV standards for
production and international program exchange, April 2002. http://www
[ITU 11] ITU-R Rec. BT.601, Studio encoding parameters of digital television for
standard 4:3 and wide screen 16:9 aspect ratios, March 2011. http://www.itu
Chapter 2. Digital Images and Video
[ITU 12] ITU-R Rec. BT.2020, Parameter values for ultra-high definition television
systems for production and international program exchange, August 2012.
[Kau 07] Kauff, P., et al., “Depth map creation and image-based rendering for
advanced 3DTV services providing interoperability and scalability,” Signal
Processing: Image Communication, vol. 22, pp. 217–234, 2007.
[Kov 14] Kovacs, P. T., A. Boev, R. Bregovic, and A. Gotchev, “Quality
measurements of 3D light-field displays,” Int. Workshop on Video Processing
and Quality metrics for Consumer Electronics (VPQM), Chandler, AR, USA,
Jan. 30–31, 2014.
[Led 13] Lederer, S., C. Muller, B. Rainer, C. Timmerer, and H. Hellwagner, “An
experimental analysis of dynamic adaptive streaming over HTTP in content
centric networks”, in Proc. of IEEE Int. Conf. on Multimedia and Expo
(ICME), San Jose, USA, July 2013.
[Mul 85] Mullen, K. T., “The contrast sensitivity of human colour vision to redgreen and blue-yellow chromatic gratings,” J. Physiol., 1985.
[Nes 67] van Nes, F. L., J. J. Koenderink, H. Nas, and M. A. Bouman, “Spatiotemporal
modulation transfer in the human eye,” Jour. of the Optical Society of America,
vol. 57, no. 9, Sept. 1967.
[MPG 07] ISO/IEC 23002-3:2007 Information technology - MPEG video
technologies - Part 3: Representation of auxiliary video and supplemental
information, 2007.
[Mul 11] Muller, K., P. Merkle, and T. Wiegand, “3-D video representation using
depth maps,” Proc. of the IEEE, vol. 99, no. 4, pp. 643–656, April 2011.
[Pan 14] Pantos, R. and W. May, HTTP Live Streaming, IETF draft-pantos-httplive-streaming-14, October 2014.
[Pin 04] Pinson, M. and S. Wolf, “A new standardized method for objectively
measuring video quality,” IEEE Trans. on Broadcasting, vol. 50, no.3,
pp. 312–322, Sept. 2004.
[Pit 13] Pitas, I., Digital Video and Television, Ioannis Pitas: 2013.
[Rei 07] Reinhard, E., T. Kunkel, Y. Marion, J. Brouillat, R. Cozot and K. Bouatouch,
“Image display algorithms for high and low dynamic range display devices,”
Jour. of the Society for Information Display, vol. 15 (12), pp. 997–1014, 2007.
[See 04] Seetzen, H., W. Heidrich, W. Stuerzlinger, G. Ward, L. Whitehead, M.
Trentacoste, A. Ghosh, and A. Vorozcovs, “High dynamic range display
systems,” Proc. ACM SIGGRAPH, 2004.
[Sha 13] Shao, F., W. Lin, S. Gu, G. Jiang, and T. Srikanthan, “Perceptual fullreference quality assessment of stereoscopic images by considering binocular
visual characteristics.” IEEE Trans. on Image Proc., vol. 22, no. 5, pp. 1940–
53, May 2013.
[Sha 98] Sharma, G., M. J. Vrhel, and H. J. Trussell, “Color imaging for multimedia,”
Proc. of the IEEE, vol. 86, no. 6, June 1998.
[Smo 11] Smolic, A., “3D video and free viewpoint video - From capture to display,”
Pattern Recognition, vol. 44, no. 9, pp. 1958–1968, Sept. 2011.
[Suc 11] Suchow, J. W. and G. A. Alvarez, “Motion silences awareness of visual
change,” Curr. Biol., vol. 21, no. 2, pp.140–143, Jan. 2011.
[Sze 11] Szeliski, R., Computer Vision: Algorithms and Applications, New York, NY:
Springer, 2011.
[Tru 93] Trussell, H. J., “DSP solutions run the gamut for color systems,” IEEE
Signal Processing Mag., pp. 8–23, Apr. 1993.
[Uka 08] Ukai, K., and P. A. Howarth, “Visual fatigue caused by viewing stereoscopic
motion images: Background, theories, and observations,” Displays, vol. 29, pp.
106–116, Mar. 2008.
[Ure 11] Urey, H., K. V. Chellapan, E. Erden, and P. Surman, “State of the art in
stereoscopic and autostereoscopic displays,” Proc. of the IEEE, vol. 99, no. 4,
pp. 540–555, April 2011.
[Wad 96] Wade, N. J., “Descriptions of visual phenomena from Aristotle to
Wheatstone,” Perception, vol. 25, no. 10, pp. 1137–1175, 1996.
[Wan 95] Wandell, B., Foundations of Vision, Sunderland, MA: Sinauer Associates,
[Wan 04] Wang, Z., A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality
assessment: From error visibility to structural similarity,” IEEE Trans. on Image
Processing, vol. 13, no. 4, pp. 600–612, Apr. 2004.
[Win 13] Winkler, S. and D. Min, “Stereo/multiview picture quality: Overview and
recent advances,” Signal Processing: Image Communication, vol. 28, no. 10,
pp. 1358–1373, Nov. 2013.
This page intentionally left blank
1D-RLC (run-length coding), 419–421
convolution summation. See Convolution
summation, 2D
image-plane motion. See Motion estimation
mesh tracking, 327–328
notation, 2
rectangular sampling, 31–32, 37–41
sampling lattices, 32–33
2D apparent-motion estimation
dense-motion estimation, 215–216
displaced-frame difference, 219–220
hierarchical motion estimation, 223–224
as ill-posed problem, 220–223
image registration, 217
optical flow equation/normal flow, 217–219
overview of, 214
performance measures for, 224–225
sparse-correspondence estimation, 214
2D apparent-motion models
defined, 210–214
non-parametric models, 213–214
parametric models, 210–213
2D-AR model, 29
2D-DCT (discrete cosine transform)
hybrid DPCM coding, 464
MC-transform coding, 466
overview of, 18–19
relationship to DFT, 20
2D-DFT (discrete Fourier transform)
boundary effects, 250
computation of, 15
DFT domain implementation, 172
diagonalizing block-Toeplitz matrices, 155
drawbacks of inverse filtering, 170
multi-frame Wiener restoration, 376
properties of, 16
2D-RLC (run-length coding), 419, 421–423
2-tap Haar filters, 445
digital cinema, 91–92
motion. See Motion estimation
steering kernel regression, SR, 394
Taylor series, 394
3D motion/structure estimation
affine reconstruction, 253–255
camera calibration, 252–253
dense structure from zero, 263
Euclidean reconstruction, 260
overview of, 251–252
planar-parallax/relative affine reconstruction,
projective reconstruction, 255–260
3D scenes, projecting onto 2D image plane,
3D video
challenges of, 85
defined, 1
disparity processing, 63
display technologies, 79–82
multi-view, 83–85
objective quality metrics for stereoscopic, 100
overview of, 79
stereoscopic video, 82–83
3D-AVC coding, 508–510
3D-DCT coding, 463–464
3D-HEVC tools, 510–512
3D-motion/pose estimation, 250
3D-transform coding, 463–466
3DTV services, 83
3D-wavelet/sub-band coding, 464–466
4 3 4 prediction, H.264/AVC, 486
16 3 8 prediction for field pictures, MPEG-2
video, 479
16 3 16 prediction, H.264/AVC, 486
24 Hz movies to 50/60 Hz conversion,
50 to 60 Hz conversion, 363
a posteriori probability. See MAP (maximum
a posteriori) probability estimates
AAC Audio, MPEG-2, 476
Above-right predictor (ARP), VSBM, 240
AC coefficients
JPEG, 435, 439–440
MPEG-1, 471–472
MPEG-1 vs. MPEG-2, 481
Accommodation distance, human stereo vision,
Active (active shutter) glasses, 80–81
Active-contour models (snakes), 287–289
Active-contour motion tracking, 325–327, 329
A/D (analog-to-digital) conversion process,
Adaptive arithmetic coding, 416, 423–424
Adaptive filters
in image enhancement, 145–146
in interpolation, 113
LMMSE, 155–157
Adaptive luminance compensation (ALC),
3D-AVV, 509
Adaptive MAP method, image segmentation,
Adaptive reference filtering coding tool, MVC,
Adaptive smoothness constraints, 232–233
Adaptive streaming
in Adobe Flash with RMTP, 93
in HTTP streaming, 93
in MPEG-DASH, 94
in Smooth Streaming, 94
in video streaming over Internet, 82
Adaptive thresholds
in change detection, 293, 295
computing, 276
in wavelet shrinkage, 160–161
Adaptive/nonlinear interpolation, 118–119
Adaptive-weighted-averaging (AWA) filter,
Additive color model, 56–57
Additive noise model, 148
Advanced residual prediction (ARP), 3D-HEVC,
Advanced Television System Committee (ATSC)
standards, 86–87
Advanced-motion vector prediction (AMVP),
HEVC, 496
AE (angular error), motion estimation, 223–224
Affine camera
in affine reconstruction, 253
orthographic projection, 199–200
paraperspective projection, 201
weak-perspective projection, 201
Affine model
in 2D-mesh tracking, 327–328
in active-contour tracking, 325
in clustering within motion-parameter, 202–203
Lukas–Kanade solution for, 229
as parametric apparent 2D-motion model,
Affine reconstruction, 253–255
Affinity-based image matting, 329
AGC (automatic gain control), 139
Aggregation, BM3D filtering, 164
ALC (adaptive luminance compensation),
3D-AVV, 509
Alias-cancellation, analysis-synthesis filters, 123,
avoiding in sampling structure conversion, 45
DFT and, 14
IIR filters in DFT domain and, 28
image decimation and, 111, 112
in LR images for SR reconstruction, 386–388
in LR images for super-resolution, 379–380
from violating Nyquist sampling rate, 39–40
Alpha-trimmed mean filters, image denoising,
Alternate scan, MPEG-2 video, 480–481
AMVP (Advanced-motion vector prediction),
HEVC, 496
Anaglyph, 80
Analog MD signal, 2
Analog video
3D sampling lattices, 33–34
analog-to-digital conversion, 33, 66–67
orthogonal sampling for progressive, 32
overview of, 63–64
progressive vs. interlaced scanning, 64–65
signal formats, 65–66
Analog-to-digital (A/D) conversion process,
Analog-video signal, 64–66
Analysis filters, 122–124, 444–445
Analysis-synthesis filters, 123, 445
Anchor pictures, MVC standard, 505
Angular error (AE), motion estimation, 223–224
Anisotropic diffusion filters, 157
Anisotropic Gaussian filtering, 119
Anti-alias filtering
down-conversion with, 351–352
down-conversion without, 352–353
in image decimation, 112–113
Aperture problem, 222–223
Apparent motion model, 206–207
Apparent-motion estimation. See 2D apparentmotion estimation
Apparent-motion models, 2D
non-parametric models, 213–214
parametric models, 210–213
Arbitrary camera pose, perspective projection,
Arbitrary slice ordering (ASO), H.264/AVC,
Arbitrary-motion trajectories, MC filtering,
Arithmetic coding
adaptive, 416
as entropy coding, 410
in image compression, 414–417
JBIG and adaptive, 423–424
ARP (above-right predictor), VSBM, 240
ARP (advanced residual prediction), 3D-HEVC,
aliasing, 116, 120
analog signal and, 64
bi-lateral filtering overcoming, 109
compression, 289, 442–443
contouring, 71
human visual system and, 58
interpolation, 118
from phase distortions in filtering, 24
regularization, 171
spatial masking and, 59
spatial resolution/frame rate and, 68
visual digital video, 96–97, 99–100
ASO (arbitrary slice ordering), H.264/AVC, 490
Aspect ratio
digital video 16:9/16:10, 76
video signal, 65
Asymmetric coding of stereo video, 506–507
Asymmetric property, DFT, 16
Asymmetric stereo-video encoding, 63
ATSC (Advanced Television System Committee)
standards, 86–87
Audio, MPEG-2, 476
Auto-calibration, camera, 252, 260
Automatic gain control (AGC), 139
Auto-stereoscopic (no glasses), 80–81, 85
Auto-stereoscopic displays, cross-talk in, 63
Average central difference, estimating partials,
AWA (adaptive-weighted-averaging) filter,
Background modeling, adaptive, 293–295
Background subtraction, change detection
adaptive background modeling, 293–295
exercises, 338
frame differencing methods, 291–293
other approaches, 297–298
overview of, 291
spatial and temporal consistency, 295
ViBe algorithm, 295–297
Backward extension of MVs, MC de-interlacing,
Backward-motion estimation, densecorrespondence, 216
circularly, 37–38
continuous signal as, 39–40
digitizing image that is not, 50
ideal spatio-temporal interpolation filter, 42
Bandpass spectral response, 54, 62
analog TV and, 65
in composite-video formats, 66
digital TV and, 86–89
HDMI full stereo-video format and, 83
in HTTP streaming, 94
multi-rate digital signal processing and, 111
required by digital video vs. audio, 71
Base view, 3D-HEVC, 510
Baseline mode, JPEG, 435–436, 438–441
Baseline profiles, H.264/AVC, 484
Baseline-depth coding, 3D-AVC, 508–509
Bayer color-filter array pattern, 69–70
Bayesian methods
image segmentation, 281–285
motion estimation, 247–249
multiple-motion segmentation, 309–311
particle-filter motion tracking, 323–325
for super-resolution reconstruction problem,
threshold values in wavelet shrinking, 161
BayesShrink, wavelets, 161
Benchmarking, segmentation/tracking results,
Berkeley segmentation dataset, 330
Bhattacharyya coefficient, mean-shift tracking,
Bi-directional prediction, MPEG-1, 472
Bi-lateral filters, 109–110, 161–162
Binarization, thresholding for, 276
Binocular depth cues, 81
Binocular disparity, 62, 80
Binocular rivalry, 63
Bi-orthogonal filters, 125–127, 446
Bit rate/quality tradeoff, JPEG, 435
Bit-depth, digital images/videos, 70–71
Bitmap parameters, 68–69
Bit-plane coding, 418–419
Bit-rates, asymmetric coding of stereo video,
Bit-stream extraction, SNR scalability, 502
BitTorrent protocol, P2PTV, 95
BLA (broken-link access) picture, HEVC, 492
Blind-image restoration, 168, 175–176
Block coding
JPEG2000, 448, 453–454
Lempel–Ziv, 430–431
overview of, 414
Block size
in basic block-matching, 134–135
in maximum-displacement estimate, 150
in motion-compensated prediction, 486–487,
in transform coding, 433
in variable-size block matching, 138–139
Block translation model, 211, 227
Blocking artifacts, 97, 464, 490
Block-matching, in motion estimation
basic procedure, 234–238
as deterministic non-parametric model,
fast search in, 236–238
full search in, 235–236
generalized, 241–242
hierarchical, 240–241
introduction to, 233–234
sub-pixel search in, 238
variable-size, 238–240
Blocks, MPEG-1, 470, 472
Block-Toeplitz matrices
in image denoising, 155
in image filtering, 168, 170
in image restoration, 167
in video filtering, 376
Block-translation motion model, 227–230
Block-wise filtering, 158, 163–164
Blue-screen matting (Chroma keying), video
capture, 329
adaptive LMMSE filter avoiding, 155
asymmetric coding and, 506–507
cross-talk causing, 63
down-conversion with anti-alias filtering and,
image decimation and, 111
image restoration undoing image, 169
image/video quality and, 96
SR in frequency domain and, 387, 389
as tradeoff in LSI denoising filter, 150
Blur identification, and blind restoration,
Blur models
overview of, 164
point-spread function, 165–167
shift-varying spatial blurring, 168
space-invariant spatial blurring, 167
Blu-ray disc specification, 361
BM3D (block-matching 3D) filtering
image de-noising with, 163–164
image restoration with, 174
V-BM3D extension for, 374
BM4D (block matching 4D) filtering, 374
Bob-and-weave filter, 357–358
in 2D recursive filtering, 28–29
of changed regions in background subtraction,
as image restoration problem, 175
in JPEG2000, 451
in phase-correlation method, 250
Box filtering, 106–108, 112
in medium-grain SNR scalability, 501
in MPEG-1, 469, 473–476
in MPEG-2, 477–478, 482–483
in temporal scalability, 498–499
in human vision, 58
in pixel-based contrast enhancement,
Broken-link access (BLA) picture, HEVC, 492
B-slices, H.264/AVC, 485, 488
Bundle adjustment, projective reconstruction,
CABAC (context-adaptive binary arithmetic
coding), H.264/AVC, 489–490, 496
Cable television
analog video format in, 64
ATSC standards for, 86
DVB-C and DVB-C2 standards for, 88–89
Calling mode, JPEG-LS, 427
Camcorders, 462
Camera calibration
in 3D motion/structure estimation, 252–253
in Dense SFS, 263
matrix, 198–199, 260
Camera projection matrices, 257
Camera shake, and image blur, 167
Camera-motion matrix, 254–255
Canny edge detection, 134–135
Cartesian coordinates, 197–198, 200
CATV, DVB-C standard for, 89
CAVLC (context-adaptive variable-length
coding), H.264/AVC, 489–490
CBs (coding blocks), 492–493
CCITT. See ITU-T (International
Telecommunications Union)
CCMF (cross-correlated multi-frame) Wiener
filter, 377
CCN (content centric networking), MPEGDASH, 94
CDF (cumulative distribution function),
138, 141
CEA (Consumer Electronics Association), 76
Center of projection, 196–199
Central difference, estimating image
gradient, 129
CFA (color filter array) interpolation, 120
Change detection, image segmentation
background subtraction, 291–298
overview of, 289
shot-boundary detection, 289–291
Change detection, in pel-recursive motion
estimation, 246
Changing elements, in 2D RLC, 421–423
Checkerboard format, stereo-video, 83
Chroma keying (blue-screen matting), video
capture, 329
composite-video format encoding, 66
perceived by human eye, 54–55
spatial-frequency response varying with,
S-video signal, 66
Chunks, HTTP, 93–94
CIE (International Commission on
Illumination), 55, 161–162
CIERGB color space, 56
CIEXYZ (or sRGB) color space, 56, 57
Circle of confusion, out-of-focus blur, 165–166
Circular convolution, 26
Circular filters, image smoothing with LSI
filter, 106
Circular shift, DFT, 17
Circular symmetry, 5, 27
Classic image/video segmentation, matting, 328
Clean random-access (CRA) picture, HEVC,
Closed GOPs, HEVC, 491–492
CLS (constrained least-squares) filtering,
170–173, 375–377
adaptive MAP method of, 284–285
for image segmentation, 277–278
K-means algorithm for, 278–279
mean-shift algorithm for, 279–280
as multiple-motion segmentation, 302–306
CMYK (Cyan, Magenta, Yellow, Black) model,
Coarse-grain SNR scalability, 501
Coder, of source encoder, 405
Coding blocks (CBs), 492–493
Coding-tree units (CTUs), HEVC, 492–495
Coefficients, DCT, 434
Coiflet filters, 124–125
Collaborative filtering, BM3D, 164
analog-video signal standards, 65–67
capture/display for digital video, 69–70
human visual system and, 54–56
image processing for digital video, 71–74
management, 57
Color balance, in image processing, 105
Color de-mosaicking, 120
Color filter array (CFA) interpolation, 120
Color space
calibrating devices to common, 57
color management and, 57
HSI (hue-saturation-intensity), 73–74
of JPEGs, 436–437
RGB and CMYK, 56–57
Color transforms, JPEG2000, 449
Color-matching functions, 54–56
Color-only pixel segmentation, 311–313
Color-region-based motion segmentation,
Color-sampling methods, image matting, 329
Complexity of content, segmentation and, 274
Component analog video, 65–67
Component color video, 67–68
Component signals, A/D conversion of, 67
Composite video, 65–67
Compositional approach, hierarchical iterativerefinement, 229
Compressed-domain methods, change detection,
image. See Image compression
video. See Video compression
Compression ratio (CR), JPEG, 442–443
Computer vision, 95
Computer/consumer electronic standards, 74, 76
Condensation (conditional-density propagation),
Conditional probability model, Bayesian image
segmentation, 282–284
Conditional replenishment (CR), 466
Cones, retinal, 54–55
Connectivity constraints, MC, 241–242
global motion, 346–347
global translation, 343–345
Constrained least-squares (CLS) filtering,
170–173, 375–377
Consumer Electronics Association (CEA), 76
Content centric networking (CCN), MPEGDASH, 94
Context, in predictive-coding mode, 427–428
Context-adaptive binary arithmetic
coding (CABAC), H.264/AVC,
489–490, 496
Context-adaptive variable-length coding
(CAVLC), H.264/AVC, 489–490
Continuous Fourier transform, 30, 34–36
Continuous signals, Fourier transform of
discrete signals vs., 12–13
overview of, 8–12
Continuous spatio-temporal intensity, optical
flow estimation, 216
Continuous-discrete model, low-resolution
sampling, 381–384
Contouring artifacts, 71
Contrast enhancement. See Enhancement, image
Contrast sensitivity, in human vision, 57–59
Contrast stretching (histogram normalization),
digital video enabling, 76
of Fourier transform, 9
MD Fourier transform of discrete signals, 12
in mean-shift clustering, 280
not an issue in DFT, 15
digitization of analog video for, 75
sampling structure for, 42–47
video-format. See Video-format conversion
Convolution summation, 2D
computation of, 22
exercise, 49
FIR filters and circular, 26
in Fourier domain, 24–25
in frequency domain analysis of sampling, 30
IIR filters and, 27–28
image smoothing with LSI filter, 106
impulse response and 2D, 21–23
Coordinate transformations, space-varying
restorations, 177
Copyrights, and digital cinema projectors, 90
Correlation coefficients, objective quality
assessment, 98
Correspondence vectors, 215–217, 242–245
Cost functions, homography, 244
Covariance-based adaptive interpolation, 119
Covered background problem, motion
estimation, 222–223
CR (compression ratio), JPEG, 442–443
CR (conditional replenishment), 466
CRA (clean random-access) picture, HEVC,
Critical velocity, 354
Cross-correlated multi-frame (CCMF) Wiener
filter, 377
Cross-talk, auto-stereoscopic displays,
63, 81
CTUs (coding-tree units), HEVC, 492–495
Cubic-convolution interpolation filter,
Cumulative distribution function (CDF),
138, 141
Data partitioning, 454, 481, 490
Data rates
3D video, 79
digital video vs. digital audio, 71
multi-view video, 83
SDTV, 75
Data structure
H.264/AVC, 484–485
HEVC, 491–492
MPEG-1, 469–471
MPEG-2 video, 477–478
Daubechies (9,7) floating-point filterbank,
JPEG2000, 450
dB (decibels), signal-to-noise ratio in, 149
DBBP (depth-based block partitioning),
3D-HEVC, 511
DC coefficient, 471–472
DC gain, 451
DC mode, 486, 495
DCI (Digital Cinema Initiative), 90
DCP (disparity-compensated prediction),
3D-HEVC, 511
DCPs (Digital Cinema Packages), 89
DCT (discrete cosine transform)
2D-DCT. See 2D-DCT (discrete cosine
3D-DCT, 463–464
image/video compression artifacts, 97
medium-grain SNR scalability and, 501
MPEG-1 and, 471–475
MPEG-2 and, 479–481
overview of, 18–19
relationship to DFT, 19–20
video compression with, 462–463
DCT coding and JPEG
encoder control/compression artifacts,
JPEG baseline algorithm, 435–436
JPEG color, 436–437
JPEG image compression, 407–408
JPEG progressive mode, 441–442
JPEG psychovisual aspects, 437–441
JPEG standard, overview, 434–435
overview of, 431–434
JPEG2000, 451–452
mid-tread uniform quantizer, 407
De-blocking filter, 490, 497
Deblurring of images. See Image restoration
Decibels (dB), signal-to-noise ratio in, 149
Decimation, image, 111–113, 117
Decoded picture buffer (DPB), H.264/AVC,
Decoder-side view synthesis, 3D-HEVC, 512
CABAC, 496
compression artifacts in, 442
H.264/AVC, 484–485, 490
HEVC parallel, 493–495
MPEG-1, 468–471, 476
MVC, 505–506
RGB sampled signal, 67
SHVC parallel, 498
into 1D transforms, 9–10
wavelet, 121–122, 126–127
Defocus blur, 96
Degradation from space-varying blurs, 177–180
Dehomogenization, perspective projection, 198
critical velocities in, 354
inter-field temporal (weave filtering), 357–358
intra-field, 355–357
motion-compensated, 359–361
overview of, 46–47
in video-format conversion, 355
De-mosaicking, color, 120
Denoising, image
image and noise models, 148–150
local adaptive filtering, 153–157
LSI filters in DFT domain, 150–153
nonlinear filtering, bi-lateral filter, 161–162
nonlinear filtering, median, 158–159
nonlinear filtering, order-statistics filters,
nonlinear filtering, wavelet shrinkage, 160–161
non-local filtering, BM3D, 163–164
non-local filtering, NL-Means, 162–163
overview of, 147–148
De-noising video. See Multi-frame noise filtering
Dense SFS (structure from stereo), 263
Dense structure from zero, 263
Dense-correspondence estimation problem, 2D,
Dense-motion (optical flow/displacement)
estimation, 215–216
Dense-motion estimation problem, 2D
correspondence vectors in, 215–216
optical flow vectors in, 216
overview of, 215
Depth coding tools, 3D-AVC, 508–509
Depth map coding
3D-HEVC, 510, 511
MVC+D (MVC plus depth maps), 507
in view-plus-depth format, 84–85
Depth perception, stereo vision, 62–63
Depth-based block partitioning (DBBP),
3D-HEVC, 511
Depth-based motion vector prediction (DMVP),
3D-AVC, 509
Depth-based representations, multi-view video,
Depth-image based rendering (DIBR), 84, 507,
Derived quantization mode, JPEG2000, 452
Deterministic non-parametric model, 213
Device-dependent color models, 56, 57
Device-independent color space, 57
DFD (displaced-frame difference)
2D apparent-motion estimation, 219–220
in Bayesian motion estimation, 247–249
in MC filter post-processing, 349
in motion-field model and MAP framework,
in pel-recursive motion estimation, 246–247
DFT (discrete Fourier transform)
in 2D DFT/inverse 2D DFT, 15–16
and boundary problem in image restoration,
convergence, 15
DCT closely related to, 19–20, 433
DCT preferred over, 18
exercise, 51
FIR filters and symmetry, 26
image smoothing with LSI filter, 106
implementing IIR filters, 28–29
implementing LSI denoising filters,
and JPEG. See JPEG standard
normalized frequency variables, 15
overview of, 14–15
in phase-correlation motion estimation, 250
properties of, 16–18
pseudo-inverse filtering with, 169
Diamond search (DS), matching method,
DIBR (depth-image based rendering), 84, 507,
Dictionary size, in Lempel–Ziv coding, 431
Difference of Gaussian (DoG) filter, 134–135
Differential image, 408
Differential methods, motion estimation
in deterministic non-parametric model, 213
Horn–Shunck method, 230–233
Lukas–Kanade method, 225–230
overview of, 225
Differential pulse-code modulation. See DPCM
(differential pulse-code modulation)
Diffusion-based-in-painting, image restoration,
Digital cinema, 89–92
Digital Cinema Initiative (DCI), 90
Digital Cinema Packages (DCPs), 89
Digital dodging-and-burning, tone mapping,
Digital images
defined, 1
finite quarter-plane support of, 3
Digital images and video
analog video, 63–67
definition of, 53
digital video. See Digital video
human vision and. See Human visual system/
overview of, 53
quality factors, 96–100
Digital micromirror devices (DMDs), DLP
projectors, 89
Digital Terrestrial Multimedia Broadcasting
standards, 86
Digital video
3D video, 79–85
analog-to-digital conversion, 33–34, 66–67
applications, 85–95
color image processing, 71–74
color/dynamic range/bit depth in, 69–71
defined, 1
Digital TV (DTV) standards, 85–89
orthogonal sampling lattice for progressive, 34
revolution in, 67
spatial resolution/frame rate in, 67–69
standards, 74–78
vertically aligned 2:1 interlace lattice
for, 34
Digital Video Broadcasting (DV) standards, 86
Digital Visual Interface (DVI) standard, 77
Digital-video applications
computer vision and scene/activity, 95
digital cinema, 89–92
digital TV (DTV), 85–89
video streaming over Internet, 92–95
Digitization, of analog, 64, 75
Direct Linear Transformation (DLT), 243–244
Direct segmentation, 307–309
Directional filtering, 20, 156–157
Directional smoothness constraints, 232–233
Discontinuity modeling, in Bayesian motion
estimation, 248–249
Discrete cosine transform. See DCT (discrete
cosine transform)
Discrete Fourier transform. See DFT (discrete
Fourier transform)
Discrete memoryless source (DMS), 402–404
Discrete random process, 402
Discrete signals
definition of, 2, 6
discrete Fourier transform (DFT), 14–18
Fourier transform of, 12–14, 35
Discrete Weiner-Hopf equation, 152
Discrete-discrete model, low-resolution
sampling, 384–386
Discrete-sine transform (DST), HEVC, 496
Discrete-trigonometric transform, DCT as, 433
Discrete-wavelet transform (DWT), 443, 448,
Disocclusion regions, 85
Disparity-compensated prediction (DCP),
3D-HEVC, 511
Displaced-frame difference. See DFD (displacedframe difference)
Display order, MPEG-1, 470–471
Display technologies
3D, 79–82
classification of, 80
digital video standards, 76
Distortion, quantizer, 406
DLP projectors, 89, 90
DLT (Direct Linear Transformation), 243–244
DMDs (digital micromirror devices), DLP
projectors, 89
DMS (discrete memoryless source), 402–404
DMVP (depth-based motion vector prediction),
3D-AVC, 509
Dodging-and-burning, image enhancement,
DoG (Difference of Gaussian) filter, 134–135
Dolby 3D cinema technology, 91
Domain filtering, 161–162, 164
Dominant-motion segmentation, 296, 299–302
Double stimulus comparison scale (DSCS), 98
Double stimulus continuous quality scale
(DSCQS), 98
Double stimulus impairment scale (DSIS), 98
with anti-alias filtering, 351–352
sampling structure conversion, 43–45
in video-format conversion, 351
without anti-alias filtering, 352–353
Down-sampling (sub-sampling)
in down-conversion, 351–354
of frame-compatible formats, 83, 503
in image decimation, 111–113
sampling structure conversion, 44
DPB (decoded picture buffer), H.264/AVC, 485
DPCM (differential pulse-code modulation)
hybrid DPCM video coding, 464
JPEG baseline mode, 436, 439
MC-DPCM video compression, 466
D-pictures, in MPEG-1, 469
avoiding in particle-filter motion
tracking, 325
in template-tracking methods, 321
DS (diamond search), matching method, 237
DSCQS (double stimulus continuous quality
scale), 98
DSCS (double stimulus comparison scale), 98
DSIS (double stimulus impairment scale), 98
DSM-CC, MPEG-2, 476
DSS (Hughes), satellite transmission, 87
DST (discrete-sine transform), HEVC, 496
DTV (Digital TV) standards, 85–89
Dual-prime prediction for P-pictures,
MPEG-2, 479
DV (Digital Video Broadcasting) standards, 86
DVB standards, 87–89
DVB-H standard, 89
DVB-S, satellite transmission, 87
DVB-S2X, 88
DVCAM format, 462
formats, 462
DVI (Digital Visual Interface) standard, 77
DWT (discrete-wavelet transform), 443, 448,
Dyadic structure, temporal scalability, 498–499
Dynamic range
compression, 143
expansion, 137
overview of, 70
Early lossless predictive coding, 424–426
EBCOT (embedded block coding with
optimized truncation), 448
EBU (European Broadcasting Union), DVB,
Edge detection
Canny, 134–135
Harris corner detection, 135–137
operators, 130–131
overview of, 127–128
Edge preserving filtering
with adaptive LMMSE filter, 155–156
bi-lateral filters, 162
with directional filtering, 156–157
with median filtering, 158–159
Edge-adaptive filters, 20, 118–119
Edge-adaptive intra-field interpolation,
Edges, modeling image, 127
Eigenvalues, Harris corner detection, 136
EM (expectation-maximization) algorithm, 174,
Embedded block coding with optimized
truncation (EBCOT), 448
Embedded quantization, JPEG2000, 452
Embedded zerotree wavelet transform (EZW),
Encoder control, 442–443, 511–512
HEVC, 493–495
HEVC parallel, 493–495
in MPEG-1, 470–471, 475–476
in MPEG-2 video, 482–483
Encrypted DCP files, 89
Endpoint error, motion estimation, 223–224
Energy-compaction property, DCT, 433–434
Enhanced-depth coding tools, 3D-AVC, 509
Enhancement, image
classifying methods of, 137
with pixel-based contrast, 137–142
with spatial filtering, 142–147
Entropy coding
H.264/AVC improvements to, 489–490
HEVC, 496
Huffman coding as. See Wavelet
JPEG 2000, 453
of symbols, 410
Entropy of source, and lossless encoding, 403
Entropy-constrained quantizers, 407
Epipolar geometry, 255–257
Equalization, histograms, 140–142
Error measures, motion estimation, 223–224
Error resilience
H.264/AVC tools for, 490–491
JPEG 2000, 454
ETSI (European Telecommunications Standards
Institute), 87–89
Euclidean reconstruction, 260
Euclidean structure, 252–254
Euler-Lagrange equations, motion estimation,
European Broadcasting Union (EBU), DVB,
Exemplar-based methods, image-in-painting,
Expectation-maximization (EM) algorithm,
174, 294
Experts, subjective quality assessments, 97–98
Expounded quantization mode, JPEG2000, 452
Extended Graphics Array (XGA), 76
Extended profiles, H.264/AVC, 484
Extrinsic matrix, perspective projection, 198
Eyes. See Human visual system/color
EZW (embedded zerotree wavelet transform),
Fast Fourier Transform (FFT) algorithms,
15–16, 18–20
Fast search, in block-matching, 236–238
Fax standards, digital, 78
Fax transmission, with RLC, 419
FD (frame difference), background subtraction,
Feature correspondences, in homography
estimation, 242–245
Feature-tracking methods, 318–321
Felzenszwalb–Huttenlocher method, graphbased video segmentation, 319
FFT (fast Fourier Transform) algorithms, 15–16,
Fidelity range extensions (FRExt), H.264/AVC,
483–484, 486, 489
Field pictures, MPEG-2, 477–480
Field prediction for field/frame pictures, MPEG2, 479
Field rate, video signal, 65, 349
Field sequential format, stereo-video
compression, 503–504
Field-DCT option in MPEG-2, 479–481, 483
Film-mode detection, frame-rate conversion, 367
Filterbanks, JPEG2000, 450
Filtered-model frame, 292
Filter-frequency response, image smoothing, 106
adaptive, 113, 145–146, 155–157
FIR. See FIR (finite-impulse response) filters
H.264/AVC in-loop de-blocking, 490
JPEG2000 normalization, 451
JPEG2000 wavelet, 450
multi-frame noise. See Multi-frame noise
video. See Video filtering
wavelet transform coding and, 443–447
zero-order hold, 115–116, 360–361
Finite differences, noise sensitivity of, 129
Finite extent signals, 2–3, 5, 14
Finite-difference operators, 128–131, 133–134
Finite-support signal, 6
FIR (finite-impulse response) filters
designing cubic-convolution filter with, 117
JPEG2000, 450
in LSI filtering for constant-velocity global
motion, 347
LSI systems, 20
and symmetry in MD systems, 25–27
wavelet representations. See Wavelet
wavelet transform coding, 445
for zero or linear phase in image processing,
Firewalls, HTT streaming and, 93
FIR-LLMSE filter, 150–151
FIR-LMMSE filter, 154–155
First derivatives, edge of 1D continuous signal,
Fixed-length coding, symbol encoding, 409–410
Fixed-reference frame, in background
subtraction, 292
Flexible macroblock ordering (FMO), H.264/
AVC, 490
Flicker frequency, and human eye, 61
Floating-point filterbanks, JPEG2000, 450
FMO (flexible macroblock ordering), H.264/
AVC, 490
analog-video signal, 65–66
conversion of video. See Video-format
digital-video spatial resolution, 68–69
stereoscopic video, 82–83
Forward prediction, MPEG-1, 472
Forward-motion estimation, 215–216
Four-fold symmetry, 5, 14, 20
Fourier transform
of continuous signals, 8–12
discrete. See DFT (discrete Fourier transform)
of discrete signals, 12–14, 35
exercises, 50
frequency response of system and, 24
impulse response of low-pass filter, 27
Nyquist criterion for sampling on lattice, 37
of signal sampled on lattice, 34–36
SR in frequency domain with, 386–387
Four-step logarithmic search, matching
method, 237
Fovea, of human eye, 54, 62
FR (full reference metrics), 98–100
Frame difference (FD), background subtraction,
Frame packing, HDMI, 83
Frame pictures, MPEG-2 video, 477–478
Frame prediction for frame pictures, MPEG-2
video, 479
Frame rate
digital video, 67–69, 90
measuring temporal-frequency response, 61
temporal scalability and, 498–499
video signal, 65
video-format conversion. See Frame-rate
conversion, video-format
Frame sequential format, stereo-video
compression, 503–504
Frame-compatible formats, stereoscopic video,
82–83, 503–504
Frame-DCT, MPEG-2, 479–481, 483
Frame-rate conversion, video-format
24 Hz movies to 50/60 Hz, 361–363
50 to 60 Hz, 363
definition of, 361
film-mode detection, 367
MC frame/scan-rate conversion, 366
motion-adaptive scan-rate conversion,
scan-rate doubling, 363–365
Frames, motion segmentation using two,
Free-view 2D video, 503
Freeze frame, 97
Frequency domain
analyzing LSI denoising filters in, 150
sampling on MD lattices, 36–41
super-resolution in, 386–389
unsharp masking image enhancement
and, 144
Frequency response
1D binary decomposition filters, 122
convolution in Fourier domain, 25
IIR Weiner filter, 152–153
MC filter, 346
MD systems, 23–25
out-of-focus blur, 166
sampling structure conversion, 44–45
of zero-order-hold filter, 115
Frequency shifting property, DCT, 19
Frequency shifting property, DFT, 17
Frequency spectrum of video, 342–345
Frequency variables, 8, 12, 14–15
FRExt (fidelity range extensions), H.264/AVC,
483–484, 486, 489
Full reference metrics (FR), 98
Full search, in block-matching, 235–236, 241
Full-resolution stereo-video format, 83
Fundamental matrix, projective reconstruction,
Games, stereo-video format for, 83
Gaussian filters
anisotropic, 119
bi-lateral filters and, 162
Canny edge detection and, 134
directional filtering using, 157
estimating Laplacian of, 134–135
estimating partials by derivatives of, 131–132
extending with bi-lateral filtering, 109–110
in image decimation, 112
image smoothing with LSI filter and, 106,
Gaussian noise, 148
Gaussian pyramid
in hierarchical motion estimation, 223–224,
JPEG hierarchical, 442
in multi-resolution frame difference analysis,
overview of, 120
Gaussians, in background modeling, 293–295
Gaze-shifting eye movements, 62
General model (depth map), 208
General motion-compensated de-interlacing,
General periodic signal, in 2D, 4
Generalized block-matching, motion estimation,
Generalized convergence, 9, 12
Geometric image formation, 196–199, 202
Ghosting, from cross-talk, 63
Gibbs potential, 315
Gibbs random field (GRF), 285–286
Glasses, stereoscopic 3D, 91
Global positioning system (GPS) motion
tracking, 317
Global thresholds, 276
Global translation, constant-velocity, 343–345
Global-MC de-interlacing, 357–358
Golomb–Rice coding, 428–429
GOP (group of pictures)
decoding in MVC, 506
HEVC, 491–492
in medium-grain SNR scalability, 501
in MPEG-1, 469–471
Gradient estimation/edge/features
Canny edge detection, 134–135
Harris corner detection, 135–137
of image gradient, 128–131
of Laplacian, 132–134
overview of, 127–128
of partials, 131–132
Gradual view refresh (GVR), 3D-AVC/MVC+D,
Graph-based methods
image segmentation, 285–287
spatio-temporal segmentation/tracking, 319
Gray codes, symbol encoding, 409–410
GRF (Gibbs random field), 285–286
Ground-truth data. See GT (ground-truth) data
Group of pictures. See GOP (group of pictures)
Groups, BM3D filtering, 164
GT (ground-truth) data
3D coordinates, Euclidean reconstruction, 260
motion vectors, motion estimation, 224–225
segmentation/tracking, 330–331
GVR (gradual view refresh), 3D-AVC/MVC+D,
H.261, 467–468
H.262, 467
H.263, 467
H.264, 88
H.264/AVC (MPEG-4 AVC/ITU-T H.264)
3D-video compression overview, 503
input-video format/data structure, 484–485
intra-prediction, 485–486
motion compensation, 486–488
MVD compression and, 507
other tools and improvements, 489–491
overview of, 483–484
stereo/multi-view video-coding extensions of,
stereo-video SEI messages in, 504
temporal scalability, 498
transform, 488–489
H.265/HEVC standard, 504, 507
Half-plane support, 2–3, 8
Harris corner detection, 135–137
HCF (highest confidence first) algorithm,
HDMI (High-Definition Multimedia Interface),
77, 83
HDR (high dynamic range), 70
HDS (HTTP Dynamic Streaming), Adobe,
HDTV (high-definition TV), 75–78, 88
Head-motion parallax, 81
Head-motion range, 81
Hermitian symmetric, 13–14, 16–17
Hessian matrix, 126
HEVC (high-efficiency video-coding) standard
3D-HEVC tools, 510–512
3D-video compression overview, 503
coding-tree units, 492–493
entropy coding, 496
intra-prediction, 495
in-loop de-blocking filter, 497
motion compensation, 495–496
motion vector coding, 496
overview of, 491
parallel encoding/decoding tools, 493–495
transform and quantization, 496
video-input format and data structure,
Hexagonal matching, 241
Hexagonal sampling, 33
HFR (high frame rate), digital video, 90
Hi10P (High 10 Profile), H.264/AVC, 484
Hi422P (High 4:2:2 Profile), H.264/AVC, 484
Hi444P (High 4:4:4 Profile), H.264/AVC, 484
Hierarchical Bayesian motion-estimation, 248
Hierarchical block matching, 234, 240–241
Hierarchical iterative-refinement, Lukas–Kanade
motion estimation, 226–230
Hierarchical mode JPEG, 442
Hierarchical motion estimation, 223–224
Hierarchical prediction structures, 498–499
High 4:2:2 Profile (Hi422P), H.264/AVC, 484
High 4:4:4 Profile (Hi444P), H.264/AVC, 484
High 10 Profile (Hi10P), H.264/AVC, 484
High dynamic range (HDR), 70
High frame rate (HFR), digital video, 90
High Profile (HP), H.264/AVC, 484
High profiles
H.264/AVC, 484, 486
MPEG-2, 482
High-Definition Multimedia Interface (HDMI),
77, 83
High-definition TV (HDTV), 75–78, 88
High-effiiciency video-coding. See HEVC (higheffiiciency video-coding) standard
Highest confidence first (HCF) algorithm,
High-level computer vision, 95
High-pass filter, wavelet analysis, 122–123
High-resolution (HR) image, super-resolution,
change detection methods, 290
normalization (contrast stretching), 137
pixel-based contrast enhancement, 138,
HLS (HTTP Live Streaming), Apple, 94
Holographic displays, 82
Homography (perspective model), 208–209, 263
Homography estimation, of motion, 229–230,
Homomorphic filtering, image enhancement,
Horizontal mode, two-dimensional RLC,
Horizontal spatial-frequency pattern, MD
signals, 6
Horizontal sum buffer (HSB), box filtering,
Horn–Shunck motion estimation, 230–233, 249
Hough transform method, 305–306
HP (High Profile), H.264/AVC, 484
HP (hi-pass) temporal bands, 465
HR (high-resolution) image, super-resolution,
HSB (horizontal sum buffer), box filtering,
HSI (hue-saturation-intensity), color image,
HTTP (hyper-text transport protocol),
streaming over, 93–94
HTTP Dynamic Streaming (HDS), Adobe, 94
HTTP Live Streaming (HLS), Apple, 94
Huber-Markov random field model, 119
Hue-saturation-intensity (HSI), color image,
Huffman coding
block, 414
in early lossless predictive coding, 424
as entropy coding, 410
Golomb–Rice coding equivalent to, 429
in image compression, 410–413
lossless coding theorem and, 404
one-dimensional RLC and, 419–421
Human visual system/color
asymmetric coding of stereo video and,
color vision and models, 54–57
computer vision aiming to duplicate, 95
contrast sensitivity, 57–59
overview of, 54
spatio-temporal frequency response, 59–62
stereo/depth perception, 62–63
Hybrid DPCM video coding, 464
Hybrid methods, MC de-interlacing, 361
Hybrid scalability, SVC compression, 502
Hyper-text transport protocol (HTTP),
streaming over, 93–94
Hysteresis thresholding, 135
IC (illumination compensation), 3D-HEVC,
ICC (International Color Consortium) profiles,
56, 57
ICM (iterated conditional mode), 283–285,
ICT (irreversible color transform), JPEG2000,
IDCT (inverse DCT), MPEG-1, 475
IDD-BM3D (iterative decouple deblurringBM3D), 174
Identity operator, regularization by CLS, 173
IDR (instantaneous decoding refresh) picture,
H.264/AVC, 491
IETF (Internet Engineering Task Force),
streaming protocols, 93
IIR (infinite-impulse response) filters, 20–21,
IIR Weiner filter, 151–153
IIR-LLMSE filter, 151–153
Ill-posed problem, 220–223
homomorphic filtering and, 146–147
retinex for removing undesirable, 146
Illumination compensation coding tool,
MVC, 506
Illumination compensation (IC), 3D-HEVC, 511
Image and video matting, 328–329
Image capture, 96
Image compression
arithmetic coding, 414–417
DCT, 432–435
definition of, 401
and digital video, 71
elements of image compression systems,
exercises, 456–459
Huffman coding, 410–414
information theoretic concepts, 402–405
ISO JPEG2000 standard, 448–454
lossless. See Lossless image compression
methods, 405–406
orthogonality in, 125
overview of, 401–402
quantization, 406–409
symbol coding, 409–410
wavelet transform coding, 443–448
Image filtering
denoising. See Denoising, image
enhancement. See Enhancement, image
exercises, 186–193
gradient estimation/edges/features. See
Gradient estimation/edge/features
overview of, 105
re-sampling/multi-resolution. See Re-sampling
restoration. See Restoration, image
smoothing, 106–110
Image formation
with affine camera, 199–201
overview of, 196
photometric effects of 3D motion, 201–202
with projective camera, 196–199
Image gradients, 176
Image in-painting, restoration, 180–181
Image matting, 329
Image models, and performance limits, 149–150
Image noise, 96
Image processing, 24, 105
Image quality. See Video/image quality
Image rectification, in Dense SFS, 263
Image registration problem, 217
Image re-sampling. See Re-sampling
Image restoration. See Restoration, image
Image segmentation
active contour models (snakes), 287–289
Bayesian methods, 277–285
clustering, 277–280
graph-based methods, 285–287
overview of, 275
thresholding, 275–277
Image sharpening. See Spatial filtering
Image smoothing. See Smoothing, image
Image- and video-compression standards, digital
video, 77–78
Imaginary part, frequency response, 24
IMAX Digital 3D cinema technology, 91–92
Impulse, MD, 6–7, 49
Impulse response
in 2D convolution, 20–23
in 2D recursive filtering, 29
of circularly symmetric low-pass filter, 27
of cubic-convolution filter, 117
of ideal interpolation filter, 115
of linear interpolation filter, 116
of MC filter, 346
in polyphase of interpolation, 117
of zero-order-hold filter, 115
Inertial sensing motion tracking, 317
Infinite-impulse response (IIR) filters, 20–21,
Information theoretic concepts, 402–405
Initialization, 296, 324
In-Loop Block-Based View Synthesis Prediction
(VSP), 3D-AVC, 509
In-loop de-blocking filter, 490, 497
Input signal, 111–114
Input-video format
H.264/AVC, 484–485
MPEG-1, 469–471
MPEG-2 video, 477–478
Instantaneous decoding refresh (IDR) picture,
H.264/AVC, 491
Integer transforms, H.264/AVC, 489
Integer-valued vectors, periodic signals, 4
Integrated Multimedia Broadcasting (ISDB)
standards, 86
Intensity, HSI and, 73–74
Interactive semi-automatic segmentation, 329
Inter-coding, in MPEG-2, 483
Inter-field line averaging, scan-rate conversion,
Inter-field temporal (weave filtering)
deinterlacing, 357–358
Inter-frame (temporal) redundancies, 461, 462
Inter-frame compression modes, MPEG-1,
Interlace lattice, 34
Interlace video input, 345
Interlaced scanning, analog-video signals, 64–65
Interlaced video, 469, 477–480
Inter-layer spatial scalable coding, 499–500
Interleaved format, stereo-video, 83
Interleaved ordering, JPEGs, 437
Intermediate signal, in image decimation,
International Color Consortium (ICC) profiles,
56, 57
International Commission on Illumination
(CIE), 55, 161–162
International Telecommunications Union. See
ITU-T (International Telecommunications
International video compression standards, 467
Internet Engineering Task Force (IETF),
streaming protocols, 93
Inter-object disparity measures, segmentation/
tracking, 330
Interoperability, with video-format conversion,
adaptive/nonlinear, 118–119
color de-mosaicking in, 120
cubic-convolution, 116–117
efficient polyphase implementation of, 117
interlacing and, 46–47
linear, 116
overview of, 113–115
sampling rate change by rational factor in,
single-frame SR in, 119
super-resolution vs. image, 379
zero-order-hold filter in, 115–116
Interpupilar distance, of average human, 62
Intersection of two lattices, 43
Inter-view coding with ALC, 3D-AVC, 509
Inter-view motion prediction, 3D-HEVC, 511
Intra-coding, MPEG-2, 483
Intra-field de-interlacing, video-format
conversion, 355–357
Intra-field line averaging, scan-rate conversion,
Intra-frame compression modes, MPEG-1,
Intra-frame image-restoration problem, 168
Intra-frame video compression, 462–463
Intra-layer spatial scalable coding, 499–500
H.264/AVC, 485–486
HEVC, 495
Inverse 2D DFT, 15–16
Inverse 2D Fourier Transform, 8, 12
Inverse 2D Fourier transform, 34–35
Inverse DCT (IDCT), MPEG-1, 475
Inverse filtering, 169–170
Inverse Fourier transform, 35
Inverse pull-down, frame-rate conversion,
Inverse quantization at decoder, JPEG2000,
Inverse wavelet transform, in image denoising,
IP multicast, 95
in MPEG-1, 469, 475
in MPEG-2 video, 477–479
Irregular Repeat-Accumulate codes, DVB-S2, 88
Irreversible color transform (ICT), JPEG2000,
I-slice, H.264/AVC, 484
ISO (International Standards Organization)
JPEG. See JPEG standard
JPEG2000. See JPEG2000 standard
video-compression standards, 467
Isometry, affine model, 212
Isomorphic signals, 4
Iterated conditional mode (ICM), 283–285,
Iterative decouple deblurring-BM3D (IDDBM3D), 174
Iterative methods, in space-varying restorations,
ITU-T (International Telecommunications
G3 and G, 419–423
H.261 standard, 467
H.264 standard
ITU-R broadcast standards, 74–77
sampling standards, 66–67
standardizing digital-image communication,
Video Quality Experts Group, 100
video-compression standards of, 467
JBIG (Joint Bi-level Image Experts Group),
Johnson filters, 125
Joint Video Team (JVT). See MPEG-4 AVC/
ITU-T H.264 standard
JP2 file format, JPEG2000, 454
JPEG standard
baseline algorithm, 435–436
color, 436–437
DCT coding and, 431–434
DV compression algorithm vs., 462–463
encoder control/compression artifacts,
as first still-image compression method, 78
hierarchical mode, 442
integer prediction in lossless mode, 425–426
lossless mode of original, 424–426
overview of, 434–435
progressive mode, 441–442
psychovisual aspects, 437–441
uniform quantization in compression,
JPEG2000 standard
boundary extension, 451
for digital cinema, 78
entropy coding, 453
error resilience, 454
filter normalization, 451
inverse quantization at decoder, 452–453
modern wavelet compression methods
influencing, 448
Motion JPEG 2000, 463
overview of, 448
prep-processing and color transforms, 449
quantization at encoder, 451–452
region-of-interest coding, 454
wavelet filters, 450
JPEG-LS standard
determination of calling mode, 427
lossless image compression with, 426
predictive-coding mode, 427–429
run-length coding mode, 429–430
JVT (Joint Video Team). See MPEG-4 AVC/
ITU-T H.264 standard
Kernel selection, directional filtering, 157
Key pictures
H.264/AVC, 485
medium-grain SNR scalability, 501
KLT (Kanade–Lucas–Tomasi) tracker, 319–321
KLT (Karhunen–Loeve transformation), 432
K-means algorithm, 278–280, 284, 302–305
K-nearest neighbor method, 280, 374
L (long) wavelength, and retinal cone, 54
Label assignment, clustering, 304–305
Labels, segmentation, 302–307
Lambertian reflectance model, 3D motion,
determining edge pixels by, 134
estimating by finite-difference operators,
of Gaussian filter, 134–135
in graph-based image segmentation, 287
JPEG2000 quantization at encoder,
in modeling image edges, 127
regularization by CLS, 173
Laplacian of Gaussian (LoG) filter, 134–135
Last-first property, Lempel–Ziv codes,
Lattice(s), MD
defining intersection of two, 43
defining sum of two, 43
definition of, 30
Nyquist criterion for sampling on, 36–41
reconstruction from samples on, 41–42
sampling on, 30–34
spectrum of signals sampled on, 34–36
LCD (liquid crystal) shutters, stereoscopic 3D
glasses, 91
LCD monitors, flicker and, 76
Least significant bit-plane (LSB), bit-plane
coding, 418
Least-squares (LS) solution, pseudo-inverse
filtering, 170
Lempel–Ziv coding, lossless image compression,
Lenticular sheets, auto-stereoscopic multi-view
displays, 80–81
HEVC, 497
MPEG-2 video, 477, 482
Lexicographic order, IIR filters, 28
Light, color models and, 56
Light field displays, 80–82
Line averaging, 356, 364
Line repetition, intra-field de-interlacing,
Linear, shift-invariant filters. See LSI (linear,
shift-invariant) filters
Linear approximations, to perspective-motion
model, 212–213
Linear contrast manipulation, pixel-based
enhancement, 139–140
Linear forward wavelet transform, image
denoising, 160
Linear interpolation filter, 116, 355–356, 379
Linear minimum mean-square error (LMMSE)
filter. See LMMSE (linear minimum meansquare error) filter
Linear phase, in image processing, 24–27
Linear shift-varying blurs. See LSV (linear shiftvarying) blurs
Linear-motion blur, 166–167, 374–375
Liquid crystal (LCD) shutters, stereoscopic 3D
glasses, 91
Live Internet broadcasting (push application), 92
Lloyd–Max quantizer, 406, 408
LMMSE (linear minimum mean-square error)
adaptive, 155–157
directional filtering, 157
IIR Weiner filter, 151–153
as optimal LSI denoising filter, 150–151
video de-noising with adaptive, 370–372
video de-noising with AWA filter vs., 373
Wiener filter giving, 173
Local adaptive filtering
adaptive LMMSE filter, 155–157
FIR-LMMSE filter, 154–155
image denoising with, 153
Local deformable motion, block-matching,
Local region-growing, image-in-painting, 181
Local thresholds, 276
LoG (Laplacian of Gaussian) filter, 134–135
Logarithmic search, matching method, 236–237
Log-likelihood, maximum-likelihood
segmentation, 306–309
Log-luminance domain, retinex, 146
Long (L) wavelength, and retinal cone, 54
Lossless image compression
adaptive arithmetic coding and JBIG,
bit-plane coding, 418–419
coding theorem, 403–404
defined, 405–406
early lossless predictive coding, 424–426
JPEG-LS standard, 426–430
Lempel–Ziv coding, 430–431
overview of, 417
RLC and ITU G3/G4 standards, 419–423
Lossy compression methods
DCT and JPEG. See DCT coding and JPEG
JBIG2 enabling for binary images, 423
lossless vs., 405
minimizing bit-rate, 404
quantization as. See Quantization
reversible color transform as, 449
transform coding as basis of standards for,
402, 431
Low-delay structure, temporal scalability,
Low-level computer vision, 95
Low-pass (LP) temporal bands, 3D-wavelet/subband coding, 465
Low-pass filters
image enhancement with unsharp masking,
KSI image smoothing, 106–109
LSI denoising filter, 150
LSI interpolation, 113–115
prior to down-conversion, 47
reconstruction from samples on lattice, 41–42
sampling structure conversion, 44
in wavelet analysis, 122–123
Low-resolution (LR) frames, 378–380, 381–386
LP (low-pass) temporal bands, 3D-wavelet/subband coding, 465
LR (low-resolution) frames, 378–380, 381–386
LS (least-squares) solution, pseudo-inverse
filtering, 170
LSB (least significant bit-plane), bit-plane
coding, 418
LSI (linear, shift-invariant) filters
in constant-velocity global motion, 346–347
convolution in Fourier domain, 24–25
denoising in DFT domain, 150–153
frequency response of system, 23–25
image smoothing with, 106–109
impulse response and 2D convolution in MD,
interpolation process with, 113–115
up-conversion of video with MC, 352–353
LSV (linear shift-varying) blurs
boundary problem in, 175
image restoration problem of, 168
overview of, 168, 169
POCS framework, 177
problem of, 168
pseudo-inverse filtering, 169–170
regularization by sparse-image modeling,
regularized deconvolution methods,
space-varying blurs, 177–180
transforming degradations into LSI
degradations, 177
Lukas–Kanade motion estimation, 225–230
contrast sensitivity in human vision, 58
in dodging-and-burning, 143
dynamic range and, 70
in MPEG-2 video, 477
perceived by human eye, 54–55
spatial-frequency response varying with,
spatio-temporal frequency eye response and,
S-video signal, 66
color model, 71–72
demonstration of JPEG-baseline algorithm,
JPEG psychovisual aspects and, 437–438
JPEG supporting, 435, 436
Luminous efficiency function, CIE, 55
M (medium) wavelength, retinal cone sensitivity
to, 54
Mach band effect, human vision, 58
Machine-learning methods, single-frame
SR, 119
Macroblocks. See MBs (macroblocks)
MAD (minimum mean absolute difference),
234–235, 241
Magnitude, frequency response, 24
Main profiles
H.264/AVC, 484
HEVC, 497
MPEG-2, 482
MAP (maximum a posteriori) probability
in adaptive/nonlinear interpolation, 119
Bayesian motion estimation, 247–249
Bayesian segmentation methods, 281–285
in blur identification from image gradients,
image restoration methods, 169
in POCS framework, 177–178
regularized deconvolution methods, 170–171
restoring images from LSI blur, 171
simultaneous motion estimation/
segmentation, 314–315
for super-resolution reconstruction problem,
in Wiener filter, 173
Mapped error value (MErrval). Golomb-Rice
coding, 429
Markov random field (MRF), adaptive image
processing, 153
Marr-Hildreth scale space theory, 108
Masking, visual, 58–59
Matching methods, motion estimation
block-matching method, 234–238
generalized block-matching, 241–242
hierarchical block matching, 240–241
homography estimation, 243–245
overview of, 233–234
variable-size block matching, 238–240
Matting, image/video, 328–329
Maximum a posteriori. See MAP (maximum
a posteriori) probability estimates
Maximum matching pel count (MPC), block
matching, 234
Maximum transfer unit (MTU), 484–485
Maximum-displacement estimate, 250
Maximum-likelihood segmentation, 306–309
Maxshift method, ROI coding, 454
m-bit Gray code, bit-plane coding, 418
MBs (macroblocks)
H.264/AVC, 484–486
HEVC coding-tree units replacing, 492–493
in MPEG-1, 469–470
in MPEG-1 encoding, 475–476
in MPEG-1 inter-frame compression,
in MPEG-1 intra-frame compression,
in MPEG-2 video, 477–482
in MVC coding tools, 506
MC (motion compensation)
with connectivity constraints, 242
H.264/AVC, 486–488
HEVC, 495–496
MC-DPCM, 466
MPEG-1, 468–469, 472–476
overview of, 466
without connectivity constraints, 241–242
MC (motion-compensated) de-interlacing,
MC (motion-compensated) filtering
arbitrary-motion trajectories and, 345–346
AWA filter, 372–373
errors in motion estimation, 347–348
fpr multi-frame noise, 369–372
frame/scan-rate conversion, 366
general MC de-interlacing, 360–361
global-MC de-interlacing, 359
LSI filtering, 346–347, 352–353
MC-LMMSE filter, 370–372
motion estimates/occlusion handling, 349
overview of, 345
reliable motion estimation, 348
in video processing, 20
MC (motion-compensated) interpolation, 47
MC zero-order hold filtering, 360–361
McCann99 retinex, 172
MCE (motion-compensation error), 223–224
MCMF (motion-compensated multi-frame)
filter, 377
MCP (motion-compensated prediction)
H.264/AVC, 485–488
in MPEG-1, 468, 472, 474
in MPEG-2 interlaced-video compression,
MC-transform coding, 466, 467
MD (multi-dimensional) sampling theory
Nyquist criterion for sampling on lattice,
overview of, 30
reconstruction from sampling on lattice,
sampling on lattice, 30–34
spectrum of signals sampled on lattice, 34–36
structure conversion, 42–47
MD (multi-dimensional) signals, 1–2, 5–8, 48
MD (multi-dimensional) systems
FIR filters and symmetry, 25–27
frequency response, 23–25
IIR filters and partial difference equations,
impulse response and 2D convolution, 20–23
overview of, 20
MD (multi-dimensional) transforms
Discrete Cosine Transform (DCT), 18–20
Discrete Fourier Transform (DFT), 14–18
Fourier transform of continuous signals, 8–12
Fourier transform of discrete signals, 12–14
overview of, 8
Mean filters, median filter vs., 159
Mean square difference (MSE), 99, 234, 241
Mean-shift (MS) algorithm, 279–280, 321–322
Mean-shift motion tracking, 321–322
Mean-square convergence, 9, 12
Mean-square quantization errors, 406, 408
Measurement matrix, multi-view Affine
reconstruction, 254–255
Median filtering
adaptive, 160
denoising using, 158–159
as energy function, 233
in motion-adaptive scan-rate conversion,
weighted, 159
Medical imaging modalities, projection slice
theorem, 12
Medium (M) wavelength, retinal cone sensitivity
to, 54
Medium-grain SNR (MGS) scalability, 501–502
Memory-management control-operation
(MMCO), H.264/AVC, 488
MErrval (mapped error value). Golomb-Rice
coding, 429
MGS (medium-grain SNR) scalability, 501–502
Mid-tread uniform quantizer, 407
Minimal-cut criterion, graph-based image
segmentation, 286–287
Minimax, 161
Minimum mean absolute difference (MAD),
234–235, 241
Mixed MD signal, 2
MJ2K (Motion JPEG 2000), 463
ML (maximum likelihood) estimate, 159,
MMCO (memory-management controloperation), H.264/AVC, 488
Modulation schemes, digital video, 87–89
Monte Carlo method, 283, 323–325
Mosaic representation (image stitching), 217
Mosquito noise, 96
Most significant bit-plane (MSB), bit-plane
coding, 418
Mother wavelet, 122
Motion (camera) matrices, 254–255
Motion blur, 96
Motion coding, 3D-HEVC, 511
Motion compensated prediction. See MCP
(motion-compensated prediction)
Motion compensation. See MC (motion
Motion detection
motion-adaptive filters using, 345
motion-adaptive scan-rate conversion and,
with successive frame differences, 292
Motion estimation
2D. See 2D apparent-motion estimation
3D motion/structure. See 3D motion/
structure estimation
3D-wavelet/sub-band coding benefits, 464
differential methods, 225–233
exercises, 268–272
image formation and, 196–201
matching methods. See Matching methods,
motion estimation
MATLAB resources, 272
MC filter in, 347–349
motion models. See Motion models
motion segmentation simultaneous with,
motion-adaptive filters not needing, 345
nonlinear optimization methods, 245–249
overview of, 195–196
performance measures, 224–225
transform-domain methods, 249–251
Motion JPEG 2000 (MJ2K), 463
Motion models
2D apparent-motion models, 210–214
apparent motion models, 206–207
overview of, 202–203
projected 3D rigid-motion models, 207–210
projected motion models, 203–206
Motion or object tracking, 274
Motion segmentation
as change detection. See Change detection,
image segmentation
dominant-motion segmentation, 299–302
motion estimation simultaneous with,
multiple-motion. See Multiple-motion
overview of, 298–299
region-based, 311–313
Motion smoothness, 247–249
Motion snake method, 325–327
Motion tracking
2-D mesh tracking, 327–328
active-contour tracking, 325–327
graph-based spatio-temporal segmentation,
Kanade–Lucas–Tomasi tracking, 319–321
mean-shift tracking, 321–322
overview of, 317–318
particle-filter tracking, 323–325
Motion trajectory, frequency spectrum of video,
Motion vector coding, HEVC, 496
Motion vectors. See MVs (motion vectors)
Motion-adaptive de-interlacing, 358
Motion-adaptive filtering, 345
Motion-adaptive noise filtering, 367–369
Motion-adaptive scan-rate conversion, 365–366
Motion-compensated (MC) de-interlacing,
Motion-compensated (MC) filtering. See MC
(motion-compensated) filtering
Motion-compensated multi-frame (MCMF)
filter, 377
Motion-compensation error (MCE), 223–224
Motion-field model, 314–315
Motion-picture industry, and Motion JPEG
2000, 463
Motion-skip mode coding tool, MVC, 506
Motion-vector field between two frames,
Moving Picture Experts Group. See MPEG
(Moving Picture Experts Group)
MPC (maximum matching pel count), block
matching, 234
MPEG (Moving Picture Experts Group)
history of, 467
inverse pull-down methods in, 363
video/audio compression with, 86
MPEG HEVC/H.265 standard, 467
MPEG-1 standard
encoding, 475–476
input-video format/data structure, 469–471
inter-frame compression modes, 472–474
intra-frame compression modes, 471–472
overview of, 468–469
quantization and coding, 474
video compression with, 78
MPEG-2 standard
for digital broadcast, 467
encoding, 482–483
H.264/AVC vs., 483–484
input-video format/data structure, 477–478
interlaced-video compression, 478–480
other tools and improvements, 480–481
overview of, 476–477
profiles and levels, 482
scalability tools introduced in, 498
temporal scalability, 498
video compression with, 78
MPEG-4 AVC and HEVC, 78
MPEG-4 AVC/ITU-T H.264 standard. See
H.264) standard
MPEG-DASH, HTTP-based streaming, 94
in MPEG-1, 471–472, 475
in MPEG-2, 481
MRF (Markov random field), adaptive image
processing, 153
MS (mean-shift) algorithm, 279–280, 321–322
MSB (most significant bit-plane), bit-plane
coding, 418
MSE (mean square difference), 99, 234, 241
MSEA (multi-level SEA), hierarchical block
matching, 240
MTU (maximum transfer unit), 484–485
Multicast streaming, 94–95
Multi-dimensional (MD) signals, 1–2, 5–8, 48
Multi-frame noise filtering
adaptive-weighted-averaging filtering,
BM4D filtering, 374
motion-adaptive noise filtering, 367–369
motion-compensated noise filtering, 369–372
overview of, 367
temporally coherent NLM filtering, 374
Multi-frame restoration, 168, 374–377
Multi-frame SR (super-resolution)
in frequency domain, 386–389
limits of, 381
modeling low-resolution sampling, 381–386
overview of, 377–378
recognition/example-based vs. reconstructionbased, 378
spatial-domain methods, 389–394
super-resolution vs. image interpolation, 379
super-resolution vs. image restoration, 379
what makes SR possible, 379–380
what super-revolution is, 378
Multi-hypothesis MCP, H.264/AVC, 488
Multi-level SEA (MSEA), hierarchical block
matching, 240
Multi-object motion segmentation, 301–302
Multi-picture MCP, H.264/AVC, 487–488
Multiple image displays, 80–81
Multiple motions, 206, 250
Multiple-motion segmentation
clustering in motion-parameter space,
MAP probability segmentation, 309–311
maximum-likelihood segmentation, 306–309
overview of, 302
Multi-resolution frame difference analysis, 293
Multi-resolution pyramid representations,
Multi-resolution representation, wavelet
decomposition as, 127
Multi-scale representation, wavelet
decomposition as, 127
Multi-view video
affine reconstruction, 254–255
compression. See Stereo/multi-view
overview of, 83–85
projective factorization, 258–259
Multi-view video coding (MVC) standard, 83,
503, 504–507
Multi-view-video-plus-depth (MVD) format,
84, 507
Mumford–Shah functional, 289
Mutual occlusion, 2D motion estimation,
MVC (multi-view video coding) standard, 83,
503, 504–507
MVC+D (MVC plus depth maps) standard,
507, 509
MVD (multi-view-video-plus-depth) format,
84, 507
MV-HEVC, 508
MVs (motion vectors)
aperture problem and, 222–223
backward extension of, 360
basic block matching, 234
displaced-frame difference method, 219–220
H.264/AVC improvements, 487
HEVC, 496
hierarchical block matching, 240
hierarchical iterative refinement, 226–227
in MPEG-2 encoding, 482–483
multi-frame restoration degraded by PSF,
in pel-recursive motion estimation, 246
in variable-size block matching, 234,
NAL (network-access layer) units
3D-HEVC, 510
H.264/AVC, 484
H.264/AVC error resilience with, 490–491
HEVC, 491
in medium-grain SNR scalability, 502
National Television Systems Committee. See
NTSC (National Television Systems
Natural codes, symbol encoding, 409–410
NDR (non-linear depth representation),
3D-AVC, 508
NE (norm of the error), motion estimation,
Near-lossless coding, predictive-coding mode,
Negative of image, in linear contrast
manipulation, 139–140
Netravali-Robbins algorithm, 246
Network-access layer. See NAL (network-access
layer) units
NLM (non-local means) filtering, image
denoising, 162–163
Node points, 2D mesh, 327–328
edge detection sensitivity to, 128
image denoising. See Denoising, image
models, 148–149, 152
multi-frame filtering of. See Multi-frame noise
variance in AWA filter, 373
as visual artifact, 96
Noiseless, Lempel–Ziv coding as, 431
Non-interleaved ordering, JPEGs, 436
Non-linear depth representation (NDR),
3D-AVC, 508
Nonlinear filtering
bi-lateral filter, 161–162
image denoising with, 158
in interpolation, 113
median filtering, 158–159
order-statistics filters, 159–160
wavelet shrinkage, 160–161
Nonlinear least-squares problem, bundle
adjustment, 259–260
Nonlinear optimization methods, 245–249
Nonlinear wavelet shrinkage, 160
Non-local filtering, 162–164
Non-local image self-similarities, sparse-image
modeling, 174
Non-local means (NLM) filtering, image
denoising, 162–163
Non-locally centralized sparse representation,
image restoration, 174
Non-normative tools, 3D-AVC and MVC+D, 509
Non-parametric model
estimating 2D motion, 213–214
Horn–Shunck motion estimation as,
mean-shift (MS) clustering as, 280
Non-rigid scene, multiple motions with possible
camera motion, 206
Non-symmetric half-plane (NSHP) support,
2–3, 28, 29
Non-symmetric half-plane symmetry, 5
Non-uniform quantization, 406
No-reference metrics (NR), 98–100
Norm of the error (NE), motion estimation,
Normal flow, OFE, 218–219
Normalization, 256, 451
Normalized DLT (normalized 8-point
algorithm), 243–244
Normalized rgb, 71–74
Normalized-cut criterion, graph-based
segmentation, 286–287
N-point DCT, 19–20
NR (no-reference metrics), 98–100
NSHP (Non-symmetric half-plane) support,
2–3, 28, 29
NTSC (National Television Systems Committee)
analog video format, 64
ATSC signals using same bandwidth of, 86
as composite video format, 66
digitization of analog video from PAL to,
n-view plus n-depth format, 503
Nyquist criterion, sampling on lattice, 36–41
Nyquist gain, JPEG2000 filter normalization,
Nyquist sampling rate, and super-resolution,
Observation noise, 170, 285
Occlusions, 85, 221–223, 349
Oculomotor mechanisms, human stereo vision,
OFE (optical-flow constraint)
differential methods of, 125–131
displaced-frame difference and,
overview of, 217–218
specifying motion field, 218–219
Open GOPs, HEVC, 491–492
Optical blurring, 164. See also Image
Optical flow
dense-motion estimation, 215–216
estimation problem, 216–219
Horn–Shunck motion estimation, 231–232
segmentation. See Motion segmentation
Optical-flow constraint. See OFE (optical-flow
Order-statistics filters, 159–160
Orthogonal filters, 124–127
Orthogonal sampling, 32–34
Orthogonal wavelet transform, 160–161
in analysis-synthesis filters, 445–446
in IIR and FIR LLMSE filters, 150–151
in image compression, 125
Orthographic projection model, 199–200
Otsu method, of threshold determination,
Out-of-focus blurs, point-spread function of,
P2P (peer-to-peer) streaming, 94–95
P2PTV networks, 95
Packetized elementary streams (PES), 85–86
PAL (Phase) format
analog video format, 64
as composite video format, 66
digitization of analog video from NTSC to,
Parallax barriers, auto-stereoscopic multi-view
displays, 80–81
Parallel encoding/decoding, HEVC, 493–495
Parallel projection, 200
Parametric model
in 2D apparent-motion, 210–213
of blur identification, 176
of dominant motion, 300
of Lukas–Kanade motion estimation, 225–230
of motion segmentation, 298–299
Paraperspective projection model, 201
Parseval’s Theorem, 18, 19
Partial camera calibration, 253, 260
Partial derivative estimation
edge detection and image gradient, 127–128
with finite-differences, 128–131
with Gaussian filter, 131–132
hierarchical iterative-refinement, 229
with Horn–Shunck motion estimation, 231
Laplacian with finite-difference operators,
Partial difference equations, and IIR filters,
Partial differential equations (PDEs), 180–181
Particle-filter motion tracking, 323–325
Pass mode, two-dimensional RLC, 421–423
Passbands, in wavelet analysis, 122–123
Passive glasses, stereoscopic multi-view displays,
Patches, 2D mesh, 327–328
Patches, non-local means filtering, 162–163
Pattern matching and substitution (PM&S),
JBIG2 encoder, 423–424
PBs (prediction blocks), 492–493, 495
PCS (Profile Connection Space), 57
PDEs (partial differential equations), 180–181
PDF (post-processing dilation filtering), 509
pdf (probability density function), 138, 247–249
Peak signal-to-noise ratio. See PSNR (peak
signal-to-noise ratio)
Pearson linear correlation coefficient, 98
Peer-to-peer (P2P) streaming, 94–95
Pel-recursive motion estimation, 245–247
Penalty parameter, effect on AWA filter, 373
Perceptual evaluation of video quality
(PEVQ), 99
Perfect-reconstruction (PR) property, 122–123
evaluating segmentation/tracking, 330–331
limits, in image denoising, 149–150
motion estimation, 224–225
quantizer, 406
Periodic boundary extension, JPEG2000, 451
Periodic signals
finite extent signals isomorphic to, 14
Fourier series coefficients as, 14
Fourier transform of discrete signals as
rectangularly, 12
as isomorphic to finite extent signals, 5
MD Fourier transform of discrete signals as,
overview of, 3–4
Periodicity matrix, 37, 48–49
Perspective projection model, with projective
camera, 196–199
PES (packetized elementary streams), 85–86
PEVQ (perceptual evaluation of video quality),
of frequency response, 24
as zero or linear in image processing, 24
zero or linear phase of FIR filters, 25–27
Phase format. See PAL (Phase) format
Phase-correlation method, motion estimation,
Photographic film, 64
Photometric effects of 3D motion, 201–202
Photoreceptors, human eye, 54
Picture types
H.264/AVC not using, 484–485
in MPEG-1, 469
in MPEG-2 video, 477
Pixel correspondences, two-view projective
reconstruction, 256
Pixel replication, with zero-order-hold filter,
pixel sampling density, 68
Pixel-based contrast enhancement
definition of, 137
histogram equalization, 140–141
histogram shaping, 141
image histogram, 138
linear contrast manipulation, 139–140
local contrast manipulation by pixel-based
operators, 141–142
overview of, 137–138
Pixel-based motion segmentation, 311–313
Pixel-based operators, contrast enhancement,
Pixel-difference, change detection, 289–290
Pixel-resolution (spatial) scalability, 499–500,
Pixels, bitmap parameters, 68–69
Pixel-wise filtering
bi-lateral filters as, 161–162
median filtering as, 158–159
NLM filtering as, 162–163
nonlinear filters as, 158
order-statistics filters as, 159–160
background subtraction with, 296
structure reconstruction, 261–263
PM&S (pattern matching and substitution),
JBIG2 encoder, 423–424
POCS (projections onto convex sets)
formulation, 177–180, 391–394
Point-spread function. See PSF (point-spread
Polarization multiplexing, stereoscopic multiview displays, 80–81
Polarizing filters, stereoscopic 3D glasses, 91
Polyphase implementation
of decimation filters, 112–113
of interpolation, 117–118
Post-processing dilation filtering (PDF), 509
in MPEG-1, 469
in MPEG-1 encoding, 475
in MPEG-1 inter-frame compression,
in MPEG-2 encoding, 482–483
in MPEG-2 video, 477–479
PR (perfect-reconstruction) property, 122–123
Pre-calibration methods, cameras, 252
Precision, JPEG standard, 434
Precision of segmentation, 274
Prediction, particle-filter motion tracking, 324
Prediction blocks (PBs), 492–493, 495
Prediction units (PUs), HEVC MV
coding, 496
Predictive-coding mode, 424, 427–429
Pre-filtering, in A/D conversion, 66–67
Prefix codes, Huffman coding, 410–413
Pre-processing, JPEG2000, 449
Prewitt operator, 130
Primary colors, mixing to create all colors, 56
Probabilistic smoothness constraints, nonparametric model, 213–214
Probability density function (pdf ), 138,
Probability distribution, of symbols, 402–403
Processing order, H.264/AVC, 485
Profile Connection Space (PCS), 57
H.264/AVC, 484
HEVC, 497
MPEG-2 video, 476–477, 482
Program stream (PS), MPEG, 86
Progressive (non-interlaced) video, 469,
Progressive conversion, interlaced to, 46
Progressive mode JPEG, 441–442
Progressive scanning, 64–65
Projected 3D rigid-motion models, 207–210
Projected motion models, 203–206
Projection slice theorem, 11–12
Projections onto convex sets (POCS)
formulation, 177–180, 391–394
Projective camera, in motion estimation,
Projective factorization, 258–259
Projective reconstruction
bundle adjustment, 259–260
multi-view projective factorization, 258–259
overview of, 255
two-view epipolar geometry, 255–257
3D-capable digital cinema video, 91
digital cinema, 89–91
discrete Fourier transform, 16–18
MD Fourier transform of continuous signals,
MD Fourier transform of discrete signals,
median filter vs. mean filter, 159
wavelet filters, 122–124
protocols, server-client streaming, 93
PS (program stream), MPEG, 86
Pseudo-inverse filtering, 169–170
PSF (point-spread function)
and blur identification from zero-crossings,
defined, 165
in image restoration and, 164
in multi-frame image restoration, 374–375
in multi-frame modeling, 375
in multi-frame Wiener restoration, 375–377
and out-of-focus blur, 165–166
P-slice, H.264/AVC, 484–485
PSNR (peak signal-to-noise ratio)
asymmetry by blurring and, 506–507
objective quality assessment with FR metric,
tradeoff in JPEG image coding, 453
Psychovisual aspects, JPEGs, 437–441
Psychovisual redundancy, 401, 405
Pull application (video-on-demand request), 92
Pull-down methods, frame-rate conversion,
Pure rotation, affine model, 211
Pure translation, affine model, 211
PUs (prediction units), HEVC MV coding, 496
Pyramid coding, JPEG hierarchical, 442
Pyramid representations, multi-resolution,
QAM modulation, 88–89
QF (quality factor) parameter, JPEG, 443
QM-encoder, 423, 424
QMF (quadrature mirror filters), 124–125
QP (quantization parameter), 489, 496
compression of images without loss of, 401
JPEG tradeoff on image size vs., 443
SNR scalability, 500–502
video. See Video/image quality
Quantitative measures, objective quality
assessment, 98–100
in analog-to-digital conversion, 66–67
H.264/AVC improvements, 489
HEVC, 496
JPEG baseline mode, 435
JPEG2000, 451–453
MPEG-1, 471, 474
MPEG-2, 480–481
noise, 96, 408–409
non-uniform, 406
uniform, 406–409
Quantization matrix, JPEG
controlling bit rate/quality tradeofff, 435
in MPEG-1, 471
psychovisual aspects, 437
scaling in JPEG, 442–443
Quantizer, of source encoder, 405
Quarter-plane support, 2–3, 7, 27
Radio frequency identification (RFID) motion
tracking, 317
RADL (random access decodable leading),
HEVC, 492
Random access, 3D-HEVC, 510
Random access decodable leading (RADL),
HEVC, 492
Random access skipped leading (RASL), HEVC,
Range filtering, 161–162
RASL (random access skipped leading),
HEVC, 492
Rate-distortion function, source-coding theorem,
Rational factor, sampling rate change by, 117–118
Raw (uncompressed) data rates, digital video,
RCT (reversible color transform), JPEG2000,
Real part, frequency response, 24
RealD 3D cinema technology, 91
Real-Time Messaging Protocol (RTMP), 93
Real-time performance, segmentation method
and, 274
Real-Time Streaming Protocol (RTSP), 93
Real-time Transport Protocol (RTP), 93
Real-valued functions, frequency response, 24
Real-valued signals, 14–18
Reciprocal lattice, 35
Recognition/example-based methods, superresolution, 378
filtering, 121
from samples on lattice, 41–42
super-resolution methods, 378
Rectangular periodic signal, 2D, 4–5, 12
Rectangular periodicity, 2D, 16
Rectangular sampling, 2D, 37, 49
Recursive filters, 28–29
Recursively computable prediction model, 424
Red, green, blue. See RGB (red, green, blue)
Reduced reference metrics (RR), 98–100
Reduced-resolution depth coding, 3D-AVC,
Redundancy reduction, 405
Reference element, two-dimensional RLC,
digital images and video, 100–103
image compression, 454–456, 459
image filtering, 181–186
MD signals and systems, 47–48
motion estimation, 263–268
video compression, 512–514, 515
video segmentation and tracking, 331–338
Reflection, Lambertian reflectance model,
Refresh rate, 61, 65, 76
Region-based motion segmentation, 311–313
Region-of-interest (ROI) coding, 454, 502
Regular mode, JPEG-LS, 426–427
operator choices, CLS, 173
in restoring images from LSI blur, 170–173
by sparse-image modeling, 173–174
Regularized deconvolution methods, 170–173
Relative address coding, two-dimensional RLC,
Relative affine structure reconstruction, 263
decimation, 111–113
gradient estimation. See Gradient estimation/
interpolation, 113–120
multi-resolution pyramid representations, 120
multi-resolution wavelet representations,
overview of, 110–111
in particle-filter motion tracking, 324
Residual planar-parallax motion model,
projected 3D rigid-motion, 209–210
independence of JPEG standard, 434
multi-frame super. See Multi-frame SR (superresolution)
reconstruction from samples on lattice, 41–42
sampling structure conversion for, 42–47
super-resolution. See Multi-frame SR (superresolution)
of volumetric display, 82
Restoration, image
blind restoration/blur identification, 175–176
blur models, 165–168
boundary problem in, 175
degradation from linear space-invariant blurs,
degradation from space-varying blurs,
image in-painting for, 180–181
overview of, 164–165
super-resolution vs., 379
Restoration, multi-frame video
cross-correlated multi-frame filter, 377
MC multi-frame filter, 377
multi-frame modeling, 375
multi-frame Wiener restoration, 375–377
overview of, 374–375
Retina, of human eye, 54
Retinex, 146, 147
Retouching, image-in-painting taken from,
Reversible color transform (RCT), JPEG2000,
RFID (radio frequency identification) motion
tracking, 317
RGB (red, green, blue) model
in color image processing, 71–73, 105
color management, 57
in component analog video, 65–66
digital color cameras and, 67–68
hue-saturation-intensity, 73–74
human eye processing of, 54–56
JPEGs, 436
normalized rgb, 73
overview of, 56
three-sensor cameras capturing, 69
Rigid scene, projected motion with static
camera in, 204–205
Rigid-motion models, projected 3D,
Ringing artifacts, 97, 442–443
RLC (run-length coding), 419–423
Roberts cross operator, gradient estimation,
Rods, sensitivity of retinal, 54
ROI (region-of-interest) coding, 454, 502
Rotation matrix, in perspective projection, 199
RR (reduced reference metrics), 98–100
RTMP (Real-Time Messaging Protocol), 93
RTP (Real-time Transport Protocol), 93
RTSP (Real-Time Streaming Protocol), 93
Run mode, JPEG-LS, 426–427
Run-length coding mode, JPEG-LS, 429–430
Run-length coding (RLC), 419–423
S (short) wavelength, retinal cone sensitivity to,
Saccadic eye movements, 62
SAD (sum of absolute differences), 236–240,
Sample preservation property, interpolation
filters, 115
3D sampling on lattices, 33–34, 50
in analog-to-digital conversion, 66–67
derivatives of Gaussian filtering, 131–132
with Fourier transform to obtain DFT, 14
low-resolution, 381–386
MD theory of. See MD (multi-dimensional)
sampling theory
rate change by rational factor, 117–118
Sampling matrix, 31
Sampling rate
in aliasing, 40
in analog-to-digital conversion, 67
causes of blur, 96
in CD-quality digital audio, 71
change by rational factor, 117–120
frame/field rate-up conversion increasing
temporal, 350
in image decimation, 112
in multi-rate digital signal processing, 111
in super-resolution, 378
what super-revolution is, 378
Sampling structure conversion, 350
Satellite television standards, 86–88
Saturation, HSI and, 73–74
of 3D-wavelet/sub-band coding, 464–466
pixel-resolution (spatial), 499–500, 502
SVC. See SVC (scalable-video coding)
temporal, 498
in wavelet analysis, 122
Scalable-video coding. See SVC (scalable-video
coding) compression
Scalar quantization, 406–409, 432
Scale-invariant feature transform (SIFT) system,
Scanning, progressive vs. interlaced, 64–65
Scan-rate doubling conversion, 363–365
SDTV (standard definition TV), 75, 78
SEA (successive elimination algorithm), 236,
SECAM (Systeme Electronique Color Avec
Memoire), 64, 66
Second derivatives, edge of 1D continuous
signal, 127–128
Sectional processing, space-varying restorations,
Segmentation. See Video segmentation and
SEI messages, in H.264/AVC and /H.265/
HEVC, 504
Self-occlusion, 2D motion estimation, 222–223
Semantically meaningful object segmentation,
Separable filters, 22–23
Separable signals, MD, 6
Sequences, MPEG-1, 469
Sequential Karhunen–Loeve (SKL) algorithm,
Server-client streaming, 92–93
Set-theoretic methods, 391–394
SFM (structure from motion) methods,
251–252. See also 3D motion/structure
SFS (structure from stereo), 251, 263
Shaping, histogram, 141
Sharpening image. See Spatial filtering
Shift of Origin, properties of DFT, 17–18
Shift-varying spatial blurring, 168
Shi–Tomasi corner detection method, 136
Short (S) wavelength, retinal cone sensitivity
to, 54
Shot-boundary detection, 289–291
SHVC, HEVC extended to, 498
SI (switching I) slice, H.264/AVC, 485
SIF (standard input format), 469
SIFT (scale-invariant feature transform), 137
Signal formats, analog video, 65–66
Signal-dependent noise, 148
Signal-independent noise, 148
Signals. See MD (multi-dimensional) signals
Signal-to-noise ratio (SNR), 149, 482
Similarity transformation, affine model, 212
Simple profile, MPEG-2, 482
Simulated annealing, MAP segmentation, 283
Simulation model encoder (SM3), MPEG-1,
Simultaneous contrast effect, 58
Simultaneous motion estimation/segmentation,
Sinc function, ideal interpolation filter, 115
Singular-value decomposition (SVD), 254–255
Skip frame, video transmission on unreliable
channels, 97
SKL (sequential Karhunen–Loeve) algorithm,
H.264/AVC, 484–486
MPEG-1, 469
SM3 (simulation model encoder), MPEG-1,
Smearing. See Blur
Smooth pursuit eye movements, 62
Smooth Streaming, Microsoft, 94
Smoothing, image
bi-lateral filtering for, 109–110
box filtering for, 107–108
Gaussian filtering for, 108–109
linear, shift-invariant low-pass filtering for,
overview of, 106
Smoothing filter, LSI denoising filter as, 150
Smoothness constraints, Horn–Shunck motion
estimation, 231–233
SMPTE (Society of Motion Picture and
Television Engineers), 66–67, 74
SMV (super multi-view) displays, 81–82
Snake method, video matting, 329
Snakes (active-contour models), 287–289
SNR (signal-to-noise ratio), 149, 482
Sobel operator, gradient estimation, 130
Source encoder, 405
Source encoding, 405
Source-coding theorem, 404–405
SP (switching P) slice, H.264/AVC, 485
Space-frequency spectral methods, motion
estimation, 251
Space-invariant spatial blurs, 167
Space-varying blurs, 177–180
Space-varying class-means, 284–285
space-varying image model, 155–156
Sparse feature matching, motion estimation, 234
Sparse modeling, single-frame SR, 119
Sparse priors, in-painting problem, 181
Sparse representations, for denoising, 160–161
Sparse-correspondence estimation, 2D, 214
Sparse-image modeling, 160–161, 173–174
Spatial (pixel-resolution) scalability, 499–500,
Spatial artifacts, 96
Spatial consistency, in background subtraction,
Spatial filtering
adaptive filtering, 145–146
definition of, 137
digital dodging-and-burning, 142–143
homomorphic filtering, 146–147
image enhancement, 142
retinex, 146
unsharp masking, 144–145
Spatial masking, 59
Spatial partial, 229, 231
Spatial profile, MPEG-2, 482
Spatial redundancy, and image compression,
Spatial resolution formats, digital-video,
Spatial resolution (picture size)
in auto-stereoscopic multi-view displays, 81
frame-compatible formats losing, 503
of video format, 349
Spatial segmentation, background subtraction
with, 295
Spatial weighting, hierarchical iterativerefinement, 229
Spatial-domain aliasing, 14, 28
Spatial-domain methods, multi-frame, 389–394
Spatial-frequency patterns
blur due to, 96
Fourier transform of continuous signals and, 8
MD signals and, 6
in spectrum of signals stamped on lattice, 35
Spatial-frequency response, human vision, 60–62
Spatio-temporal filtering
frequency spectrum of video, 342–345
motion-adaptive filtering, 345
motion-compensated filtering, 345–349
overview of, 342
Spatio-temporal intensity pattern, video as, 53
Spatio-temporal median filtering, 365–366
Spatio-temporal resolution
interlacing and, 47
reconstruction from samples on lattice, 41–42
sampling structure conversion for, 42–47
Spatio-temporal segmentation, 274
Spearman rank-order correlation coefficient, 98
Special multidimensional signals, 5
Spectral redundancy, 401
Spectrum of signals, sampled on lattice, 34–36
SPIHT (set partitioning in hierarchical trees),
SPM (soft pattern matching) method, 423–424
Sports video, shot-boundary detection
challenges, 290
SR (super-resolution). See Multi-frame SR
sRGB (CIEXYZ) color space, 56, 57
SSD (sum of squared differences), 135–136, 512
SSIM (structural similarity index), 99
Stability, testing for 2D recursive filtering, 28
Stable filters, 21–22
Standard definition TV (SDTV), 75, 78
Standard input format (SIF), 469
adaptive streaming, 94
analog-video, 65–66
digital cinema, 89–90
digital-video, 73–74
image/video compression, 78
sampling parameter, 66–67
sampling structure conversion, 42–47
subjective quality assessments, 97–98
Static camera, projected motion with, 203–204
Static scene, projected motion in, 203–204
Static-volume 3D displays, 82
Statistical redundancy, data compression by, 405
Stein’s unbiased risk estimate (SURE), 161
Step-size, uniform quantizer, 407
Stereo, dense structure from, 263
Stereo vision, 62–63, 203–204
Stereo/depth perception, 62–63
Stereo-disparity estimation, 216
Stereo/multi-view compression
frame-compatible formats, 503–504
MVD extensions, 507–512
overview of, 502–503
video coding extensions of H.264/AVC,
Stereoscopic (with glasses), 80–83, 91, 100
Stereoscopy, creating illusion of depth, 62–63
Stereo-video information (SVI) messages,
H.264/AVC, 504
Still images, 53, 60
Storage, MPEG-2 built for DVD, 78
Stream switching, 92
Structural similarity index (SSIM), 99
Structure from motion (SFM) methods,
251–252. See also 3D motion/structure
Structure from stereo (SFS), 251, 263
Sturm–Triggs method, 259
Sub-band coding
3D-wavelet and, 464–466
DWT and, 443–447
wavelet image coding vs., 447–448
wavelet-transform coding and, 443
Sub-pixel displacement, for super-resolution,
Sub-pixel motion estimation, SR without, 394
Sub-pixel search, block-matching, 238
Sub-sampling. See Down-sampling (subsampling)
Subtractive color model, 56–57
Successive elimination algorithm (SEA), 236,
Successive iterations, regularization by CLS,
Sum of absolute differences (SAD), 236–240,
Sum of squared differences (SSD), 135–136, 512
Super multi-view (SMV) displays, 81–82
Super multi-view video, 83–85
Super voxels, graph-based video segmentation,
Super-resolution. See Multi-frame SR (superresolution)
Suppression theory of stereo vision, 63
SURE (Stein’s unbiased risk estimate), 161
SureShrink, wavelets, 161
SVC (scalable-video coding) compression
benefits of, 497–498
hybrid scalability, 502
quality (SNR) scalability, 500–502
spatial scalability, 499–500
temporal scalability, 498–499
SVC (scalable-video encoding) compression, 92
SVD (singular-value decomposition), 254–255
SVI (stereo-video information) messages, H.264/
AVC, 504
S-video (Y/C video), formats, 65–66
Sweet spots, auto-stereoscopic multi-view
displays, 81
Swept-volume 3D displays, 82
Switching I (SI) slice, H.264/AVC, 485
coding in image compression, 409–410
information content of, 402–403
Symlet filters, 124–125
Symmetric block-matching, MC up-sampling,
Symmetric boundary extension, JPEG2000, 451
Symmetric filters, 26–27, 125–127
Symmetric signal, 5, 20
analysis-synthesis filtering property, 123
in analysis-synthesis filters, 445
as DCT property, 18–19
as DFT property, 16
FIR filters in MD systems and, 25–27
MD Fourier transform of discrete signals as, 13
properly handling image borders with, 125
Synthesis filters, wavelet analysis, 122–124
Systeme Electronique Color Avec Memoire
(SECAM), 64, 66
Systems, MPEG-2, 476
TBs (transform blocks), HEVC, 493, 495
TCP (transmission control protocol), serverclient streaming, 92–93
Television. See TV (television)
Template-tracking methods, 318, 321
Temporal (inter-frame) redundancies, 461, 462
Temporal artifacts, 96
Temporal consistency, 295, 301
Temporal masking, 59
Temporal partial, 229, 231
Temporal prediction, MPEG-1, 472
Temporal scalability, 498–499, 502
Temporal segmentation, 289
Temporal-frequency, 61–62, 342–343
Temporally coherent NLM filtering, 374
Temporal-prediction structures, H.264/AVC,
Terrestrial broadcast standards, 86–87
Terrestrial mobile TV broadcasts, DVB-H for, 89
Test Model 5 (TM5) encoder, MPEG-2,
Test Zone Search (TZZearch) algorithm,
Testing, and subjective quality assessments,
Texture coding tools, 3D-AVC, 509
Texture mapping, 2D-mesh tracking, 327–328
Three-step search (TSS), matching method,
determining edge pixels by, 134
estimating in wavelet shrinking, 160–161
finding optimum threshold, 276–277
for image segmentation, 275–276
in multi-resolution frame difference
analysis, 293
Tier-1 coding, JPEG 2000 operation, 448
Tier-2 coding, JPEG 2000 operation, 448
Tiers, HEVC, 497
Tiles, HEVC, 493–494
Tiles, JPEG2000, 448, 449
Time-first decoding, MVC, 505–506
Time-recursive (TR) filters, MC de-interlacing,
Time-sequential sampling, 33–34
TM5 (Test Model 5) encoder, MPEG-2,
Tone mapping. See Pixel-based contrast
enhancement; Spatial filtering
TR (time-recursive) filters, MC de-interlacing,
Transform blocks (TBs), HEVC, 493, 495
Transform coding
DCT. See DCT (discrete cosine transform)
DCT vs. H.264/AVC, 488–489
H.264/AVC, 488–489
HEVC, 496
MD Fourier transform. See MD (multidimensional) transforms
Transform-domain methods, motion estimation,
Transmission control protocol (TCP), serverclient streaming, 92–93
Transport stream, ATSC, 86–87
Transport stream (TS), MPEG, 86
Tree diagram, Huffman coding, 412
Triangulation, projective reconstruction, 257
Trimap (tri-level image segmentation), 328
Tri-stimulus theory, digital color, 56
Tri-stimulus values, color vision and, 54
TS (transport stream), MPEG, 86
TSS (three-step search), matching method,
TV (television)
analog, 64
broadcast standards, 74–76
digital TV (DTV), 85–89
frame rate for, 61
frame rate standards, 361
frame-compatible stereo-video formats for
3DTV, 83
intra-frame compression in, 462
scan-rate doubling used in, 363
Two-field median filter, motion-adaptive
de-interlacing, 358
Two-fold symmetry
definition of, 5
in Fourier domain, 14
impulse response of low-pass filter, 27
MD Fourier transform of discrete
signals, 14
signals, 5
Two-level binary-tree decomposition, 126
Two-step iteration algorithm, 316–317
Two-view epipolar geometry, projective
reconstruction, 255–257
Types I-IV DCT, 18–19
Types V-VIII DCT, 18
TZZearch (Test Zone Search) algorithm,
UDP (User Datagram Protocol), 92–93, 95
UHDTV (ultra-high definition television),
76, 78
Ultra-high definition television (UHDTV),
76, 78
UMHexagonS, matching method, 237–238
Uncompressed (raw) data rates, digital video,
Uncovered background, in motion estimation,
Uniform convergence, 9, 12
Uniform quantization
defined, 406
in JPEG2000, 448, 451–452
with Lloyd–Max quantizers, 407
in MPEG-1, 471
overview of, 407–409
Uniform reconstruction quantization (URQ),
HEVC, 496
Uniform velocity global motion, 344
Unit step, MD signals, 7–8
Unitless coordinates, 32
Unit-step function, modeling image edges, 127
Unsharp masking (USM), image enhancement,
Up-conversion, sampling structure conversion,
URQ (uniform reconstruction quantization),
HEVC, 496
User Datagram Protocol (UDP), 92–93, 95
USM (unsharp masking), image enhancement,
Variable-block-size MCP, H.264/AVC, 487
Variable-length source coding. See VLC
(variable-length source coding)
Variable-size block matching (VSBM), 234,
Variance of noise, SNR, 149
VBV (video buffer verifier), MPEG-1, 476
VCEG (Video Coding Experts Group), ITU-T,
Vector quantization, 406
Vector-field image segmentation, 285
Vector-matrix model, 200, 375
2D sampling lattices, 32
3D sampling lattices, 33–34
MD sampling lattice, 30–31
Vergence distances, human stereo vision, 62–63
Vertical mode, two-dimensional RLC, 421–423
Vertical resolution, video signals, 65
Vertical spatial-frequency pattern, MD
signals, 6
Vertical sum buffer (VSB), box filtering,
VESA (Video Electronics Standards
Association), 76
VGA (Video Graphics Array) display
standard, 76
ViBe algorithm, 295–297
MD signals/systems in, 1
as spatio-temporal intensity pattern, 53, 61–62
temporal-frequency response in, 61
Video buffer verifier (VBV), MPEG-1, 476
Video Coding Experts Group (VCEG), ITU-T,
Video compression
3D-transform coding, 463–466
digital TV, 85–86
fast search, 236–238
intra-frame, 462–463
motion-compensated transform coding, 466
overview of, 461–462
scalable video compression, 497–502
stereo/multi-view, 502–512
Video compression standards
digital-video, 74–78
HEVC, 491–497
international, 467
ISO and ITU, 467–468
Motion JPEG 2000, 463
MPEG-1, 468–476
MPEG-2, 476–483
MPEG-4/ITU-T H.264, 483–491
Video Electronics Standards Association (VESA),
Video filtering
multi-frame noise filtering, 367–374
multi-frame restoration, 374–377
multi-frame SR. See Multi-frame SR (superresolution)
overview of, 341
spatio-temporal filtering theory, 342–349
video-format conversion. See Video-format
Video Graphics Array (VGA) display standard,
Video matting methods, 329
Video Quality Experts Group (VQEG), 100
Video segmentation and tracking
change detection. See Change detection,
image segmentation
exercises, 239
factors affecting choice of method for, 274
image and video matting, 328–329
image segmentation. See Image segmentation
motion segmentation. See Motion
motion tracking. See Motion tracking
overview of, 273–275
performance evaluation, 330–331
Video streaming over Internet, 92–95
Video view coding, 3D-HEVC, 510
Video-format conversion
de-interlacing, 355–361
down-conversion, 351–355
frame-rate conversion, 361–367
overview of, 349–350
problems of, 350–351
Video-format standards, digital video, 74–77
Video/image quality
objective assessment of, 98–100
overview of, 96
subjective assessment of, 97–98
visual artifacts, 96–97
Video-interface standards, digital video, 77
Video-on-demand request (pull application), 92
View synthesis, 3D-HEVC, 510
View synthesis distortion (VSD), 509
View synthesis prediction coding tool, MVC,
View synthesis prediction (VSP), 509, 511
View-first decoding, MVC, 505–506
Viewing distance, human eye response, 60
Viewing-position-dependent effects, volumetric
displays, 82
View-plus-depth format, in multi-view video,
Vision persistence, 61
Visual artifacts, 96–97
Visual masking, human vision, 58–59
Visual motion tracking, 317–318
Visual-quality degradation (loss), image
compression, 401
VisuShrink, 161
VLC (variable-length source coding)
defined, 402
as entropy coding, 410
JPEG baseline mode, 435–436
in MPEG-1, 468, 471–472, 474–475
Volumetric displays, 80, 82
Voronoi cell of a 2D lattice, 36
Voxels, 82
VQEG (Video Quality Experts Group), 100
VQM, objective quality assessment, 99–100
VSB (vertical sum buffer), box filtering, 107–108
VSBM (variable-size block matching), 234,
VSD (view synthesis distortion), 509
VSP (view synthesis prediction), 509, 511
White noise
adaptive LMMSE filter and, 155–156
defined, 148
IIR Weiner filter and, 153
image denoising with wavelet shrinkage, 160
Wiener filter
collaborative, 164
deconvolution filter, 170–171, 173
IIR (infinite-impulse response), 151–153
image restoration with, 169
multi-frame restoration, 375–377
Wiener-based motion estimation, 147
Warping, hierarchical iterative-refinement, 229
Wavefront parallel processing, HEVC, 494–495
Wavelengths, color sensitivity and, 54–56
Wavelet filters, JPEG2000, 450
Wavelet representations
with bi-orthogonal filters, 125–127
in image re-sampling, 121–124
with orthogonal filters, 124–125
Wavelet shrinkage, image denoising, 160–161
Wavelet transform coding
choice of filters, 443–447
overview of, 443
sub-band vs. wavelet image coding, 447–448
wavelet compression, 448
Weak-perspective projection model, 200
Weave filtering (inter-field temporal)
deinterlacing, 357–358
Weber’s law, 58
Wedge support, MD signals, 2–3
Weighted MCP, H.264/AVC, 488
Weighted median filtering, image denoising, 159
Weights for patches, NL-means filtering,
Y/C video (S-video), formats, 65–66
Y-Cr-Cb color space
in color image processing, 71–74, 105
JPEG, 436
MPEG-1, 469
x264 library, H.264/AVC, 491
x265 library, HEVC, 497
XGA (Extended Graphics Array), 76
XpanD 3D cinema technology, 91
Zero Fourier-phase, 14
blur identification from, 175–176
linear motion blur, 167
multi-frame restoration and, 275
out-of-focus blur, 166
super-resolution in frequency domain and, 288
Zero-mean noise, 155–156
Zero-order coder, 426
Zero-order hold filtering, 115–116, 360–361
Zero-phase filters, 25–27, 445
Zigzag scanning
in MPEG-2, 480
of quantized AC coefficients, 439, 472
of quantized DCT coefficients, 435, 438,
441, 474
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF