IMAGINE DeltaCue - University of New England

IMAGINE DeltaCue
User’s Guide
September 2008
Copyright © 2008 ERDAS, Inc.
All rights reserved.
Printed in the United States of America.
The information contained in this document is the exclusive property of ERDAS, Inc. This work is protected under
United States copyright law and other international copyright treaties and conventions. No part of this work may be
reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying and
recording, or by any information storage or retrieval system, except as expressly permitted in writing by ERDAS, Inc.
All requests should be sent to the attention of:
Manager, Technical Documentation
ERDAS, Inc.
5051 Peachtree Corners Circle
Suite 100
Norcross, GA 30092-2500 USA.
The information contained in this document is subject to change without notice.
Government Reserved Rights. MrSID technology incorporated in the Software was developed in part through a
project at the Los Alamos National Laboratory, funded by the U.S. Government, managed under contract by the
University of California (University), and is under exclusive commercial license to LizardTech, Inc. It is used under
license from LizardTech. MrSID is protected by U.S. Patent No. 5,710,835. Foreign patents pending. The U.S.
Government and the University have reserved rights in MrSID technology, including without limitation: (a) The U.S.
Government has a non-exclusive, nontransferable, irrevocable, paid-up license to practice or have practiced
throughout the world, for or on behalf of the United States, inventions covered by U.S. Patent No. 5,710,835 and has
other rights under 35 U.S.C. § 200-212 and applicable implementing regulations; (b) If LizardTech's rights in the
MrSID Technology terminate during the term of this Agreement, you may continue to use the Software. Any provisions
of this license which could reasonably be deemed to do so would then protect the University and/or the U.S.
Government; and (c) The University has no obligation to furnish any know-how, technical assistance, or technical data
to users of MrSID software and makes no warranty or representation as to the validity of U.S. Patent 5,710,835 nor
that the MrSID Software will not infringe any patent or other proprietary right. For further information about these
provisions, contact LizardTech, 1008 Western Ave., Suite 200, Seattle, WA 98104.
ERDAS, ERDAS IMAGINE, IMAGINE OrthoBASE, Stereo Analyst and IMAGINE VirtualGIS are registered trademarks;
IMAGINE OrthoBASE Pro is a trademark of ERDAS, Inc.
SOCET SET is a registered trademark of BAE Systems Mission Solutions.
Other companies and products mentioned herein are trademarks or registered trademarks of their respective owners.
iii
iv
Table of Contents
Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
About This Users Guide . . . . . . . . . . . . . . . . . . . . . . . . 1
Example Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1
Conventions Used in This Book
Terminology . . . . . . . . . . . . . . .
Special Characters and Fonts . . .
Special Paragraphs . . . . . . . . . .
. . . . . . .
........
........
........
.
.
.
.
. . . . . . .
........
........
........
.
.
.
.
. .1
.. 1
.. 2
.. 2
Getting Started with DeltaCue . . . . . . . . . . . . . . . . . . . . . . 3
ERDAS IMAGINE Icon Panel . . . . . . . . . . . . . . . . . . . . . 3
Help System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Preference Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Process Inputs/Outputs . . . . . . . . . . . . . . . . . . . . . . . 5
Change Detection Using DeltaCue . . . . . . . . . . . . . . . . . . . . 7
What is DeltaCue? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Change Detection Basics . . . .
Preliminary Steps . . . . . . . . . .
Change Detection Methods . . .
Filtering Unwanted Change . . .
Analyzing Change . . . . . . . . . .
.
.
.
.
.
. . . . . . .
........
........
........
........
.
.
.
.
.
. . . . . . .
........
........
........
........
.
.
.
.
.
.
.
.
.
.
DeltaCue Change Detection .
Processing Strategy . . . . . . .
Pre-Processing . . . . . . . . . . .
Change Algorithms . . . . . . . .
Change Threshold . . . . . . . . .
Change Filtering . . . . . . . . . .
Change Results Viewing . . . . .
Process Management . . . . . .
.
.
.
.
.
.
.
.
. . . . . . .
........
........
........
........
........
........
........
.
.
.
.
.
.
.
.
. . . . . . .
........
........
........
........
........
........
........
.
.
.
.
.
.
.
.
. 23
. 23
. 24
. 25
. 30
. 30
. 32
. 32
.
.
.
.
.
.
.
.
.9
10
12
16
22
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
DeltaCue Tutorial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Data Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Initial Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Change Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Iteration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
v
DeltaCue Wizard Interface . . . . . . . . . . . . . . . . . . . . . . . . 53
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Initial Wizard Usage . . . . . . . . . . . . . . .
Project Selection Dialog . . . . . . . . . . . . . . .
Image Cropping Dialog . . . . . . . . . . . . . . .
Normalization Dialog . . . . . . . . . . . . . . . . .
Change Detection Dialog . . . . . . . . . . . . . .
Change Filters Dialog . . . . . . . . . . . . . . . .
Material Filter Dialog . . . . . . . . . . . . . . . . .
Session Output . . . . . . . . . . . . . . . . . . . . .
Setting Interactive Change Thresholds . . . .
.
.
.
.
.
.
.
.
.
. . . . . . .
........
........
........
........
........
........
........
........
.
.
.
.
.
.
.
.
.
. 53
. 54
. 55
. 56
. 57
. 60
. 63
. 64
. 65
Wizard Usage with Existing Project Files . . . . . . . . . . 67
DeltaCue Change Display Viewer . . . . . . . . . . . . . . . . . . . . 69
Viewer Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Viewer Menus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Viewer Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
DeltaCue Iterations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Iteration Dialog . .
Change Algorithms
Change Filters . . .
Material Filters . . .
Saving Settings . .
. . . . . . . .
.........
.........
.........
.........
.
.
.
.
.
. . . . . . .
........
........
........
........
.
.
.
.
.
. . . . . . .
........
........
........
........
.
.
.
.
.
. 81
. 83
. 85
. 87
. 88
Iteration Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
DeltaCue Site Monitoring Mode . . . . . . . . . . . . . . . . . . . . . 91
Site Monitoring Process . . . . . . . . . . . . . . . . . . . . . . . 91
Create a New Project . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Use an Existing Project . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Site Monitoring Viewer . . . . . . . . . . . . . . . . . . . . . . . 93
Material View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Multitemporal View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
DeltaCue Workspaces . . . . . . . . . . . . . . . . . . . . . . . . . . 101
DeltaCue Material Filtering . . . . . . . . . . . . . . . . . . . . . . . 105
Material Filter Parameters . . . . . . . . . . . . . . . . . . . . 105
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
vi
List of Figures
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
1: Co-registration of a slave image (blue) to the master image (green).
2: Histogram of a change image. . . . . . . . . . . . . . . . . . . . . . . . . . .
3: Histogram of a change image with thresholds . . . . . . . . . . . . . . . .
4: Histogram of a change image with bounding thresholds. . . . . . . . . .
5: Change Region Blob . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6: Properly and improperly registered image pairs. . . . . . . . . . . . . . .
7: Representation of the effect of misregistration. . . . . . . . . . . . . . .
8: Detection of Misregistration. . . . . . . . . . . . . . . . . . . . . . . . . . . .
9: Spectral segmentation using unsupervised classification. . . . . . . . .
10: Tutorial Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11: Time 1 Image and Histogram . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 11
. 14
. 15
. 16
. 17
. 19
. 19
. 20
. 21
. 36
. 44
vii
viii
List of Tables
Table 1: Comparison of Compactness Values for Various Rectangular Shapes . . . . . . . . 18
Table 2: Tutorial Image Data Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Table 3: Geometric Properties Used for Spatial Filtering . . . . . . . . . . . . . . . . . . . . . . 62
ix
x
Preface
About This Users
Guide
This Users Guide is designed to aid you in understanding the
effective use of the DeltaCue software for image pair change
detection tasks. It provides a general introduction to the topic of
change detection in remote sensing and details the procedures
implemented in the initial release of the DeltaCue software. The
Users Guide steps you through the processes, including inputs, preprocessing, change detection, thresholding, results filtering and
display. It also gives you tips on the proper implementation of the
techniques and presents knowledgeable advice for attaining the best
results using the software.
Example Data
Sample data sets are provided with the software. This data is
separately installed from the data DVD. For the purposes of
documentation, <ERDAS_Data_Home> represents the name of the
directory where sample data is installed. The Tour Guides refer to
specific data which are stored in <ERDAS_Data_Home>/examples.
Documentation
This manual is part of a suite of on-line documentation that you
receive with ERDAS IMAGINE software. There are two basic types of
documents, digital hardcopy documents which are delivered as PDF
files suitable for printing or on-line viewing, and On-Line Help
Documentation, delivered as HTML files.
The PDF documents are found in <IMAGINE_HOME>\help\hardcopy
where <IMAGINE_HOME> represents the name of the directory in
which ERDAS IMAGINE is installed. Many of these documents are
available from the ERDAS Start menu. The on-line help system is
accessed by clicking on the Help button in a dialog or by selecting an
item from a Help menu.
Conventions Used
in This Book
In ERDAS IMAGINE, the names of menus, menu options, buttons,
and other components of the interface are shown in bold type. For
example:
“In the Raster dialog, select the Fit to Frame option.”
Raster is the name of a specific dialog and Fit to Frame is an option
within that dialog.
Terminology
Conventions Used in This Book
When asked to use the mouse, you are directed to click, shift-click,
middle-click, right-click, hold, drag, etc.
•
click — designates clicking with the left mouse button
•
shift-click — designates holding the Shift key down on your
keyboard and simultaneously clicking with the left mouse button
•
middle-click — designates clicking with the middle mouse button
1
Special Characters and
Fonts
•
right-click — designates clicking with the right mouse button
•
hold — designates holding down the left (or right, as noted)
mouse button
•
drag — designates dragging the mouse while holding down the
left mouse button
Certain characters and font styles have special meaning in ERDAS
IMAGINE documentation.
When you see words enclosed in < >, substitute the proper
information for these words. For example, when you see
<IMAGINE_HOME> in this document, replace it with the name of the
drive and directory where ERDAS IMAGINE is installed on your
system.
Special Paragraphs
The following paragraphs are used throughout ERDAS IMAGINE
documentation:
These paragraphs contain strong warnings or important tips.
These paragraphs direct you to the ERDAS IMAGINE software
function that accomplishes the described task.
These paragraphs lead you to other sections of this manual or
other specified manuals for additional information.
These paragraphs draw your attention to useful information.
NOTE: Notes give additional instruction.
2
Conventions Used in This Book
Getting Started with DeltaCue
The DeltaCue software package is fully integrated with ERDAS
IMAGINE software allowing you to make use of IMAGINE’s advanced
viewing and processing functionality for all DeltaCue process inputs
and outputs. A wizard-type interface steps you through the initial
change detection process and a special DeltaCue change display
viewer with features specifically designed for change detection
analysis may be used to view the results. From the change display
viewer, you can then execute iterations on the original process until
you get the exact results that you want and then save those settings
for future processing runs. The combination is a powerful operational
tool for detecting change in image pairs acquired by spectral imaging
sensors.
ERDAS IMAGINE
Icon Panel
DeltaCue software is installed as a main process within the IMAGINE
processing suite. Access its functionality by clicking the DeltaCue
icon on the main IMAGINE icon panel or by selecting Main >
DeltaCue... in the IMAGINE Main menu.
Once you have elected to start the DeltaCue process, the main
DeltaCue dialog is displayed. Refer to “DeltaCue Wizard Interface” on
page 53 for detailed instructions on how to use the DeltaCue wizard
to get started processing.
Help System
Help System
There are several ways to obtain more information regarding
dialogs, tools, or menus, as described below.
3
On-Line Help
There are two main ways you can access On-Line Help in ERDAS
IMAGINE:
•
select the Help option from a menu bar
•
click the Help button on any dialog
Status Bar Help
The status bar at the bottom of the Viewer displays a quick
explanation for buttons when the mouse cursor is placed over the
button. It is a good idea to keep an eye on this status bar, since
helpful information displays here, even for other dialogs.
Bubble Help
The User Interface and Session category of the Preference Editor
enables you to turn on Bubble Help, so that the single-line Help
displays directly below your cursor when your cursor rests on a
button or frame part. This is helpful if the status bar is obscured by
other windows.
Preference
Settings
Before using DeltaCue software within IMAGINE, it is best to change
certain IMAGINE preference settings to maximize ease of use.
It is recommended that you set the following preference settings
under the User Interface & Session section as described below.
1. From the main menu, select Session | Preferences. The
Preference Editor dialog opens.
2. Click in the category User Interface & Session to display the
specific options for this category.
3. Locate the option Keep Job Status Box as shown here, and
uncheck the checkbox.
The DeltaCue process runs a number of separate job tasks during
processing. If this preference is not turned off, you will have to select
OK to dismiss each job status dialog when the process has
completed.
4
Preference Settings
4. In the same preference category, you may also wish to check Use
Bubble Help.
5. Under the Preference Category of Image Files (General) the
Compute Pyramid Layers should be checked. If left unchecked the
software will prompt you for the calculation of pyramid files when it
displays the imagery.
6. Click the User Save button to save your changes.
Process
Inputs/Outputs
The DeltaCue process is designed to work with uncalibrated
imagery in IMAGINE .img, .ntf, .tif, and .jp2 formats. The image
data need not be corrected for atmospheric effects since sceneto-scene normalization is part of the DeltaCue process. Images
in other image formats need to be converted to .img format prior
to use with the DeltaCue process.
The DeltaCue process produces a color-coded change detection
output layer that is stored in a raster .img file. This image is overlaid
on the original image in the DeltaCue change display viewer and may
also be viewed using the regular ERDAS IMAGINE viewer. Other
IMAGINE tools can then be used to produce output products. The
displayed data can be saved to a wide range of standard output
formats including JPEG and GeoTIFF. You can also use the ERDAS
IMAGINE annotation capability to add annotation to your data and
produce map compositions for final output.
Process Inputs/Outputs
5
The DeltaCue process produces a single final output change
detection layer which represents the composite detection of all
significant change between the two images specified. Individual
intermediate change detection files and masks are retained in the
project workspace directory. These files are also stored in IMAGINE
.img format and can be viewed and used to produce output products.
Refer to “DeltaCue Workspaces” on page 101 for more information
about these outputs.
The output of the DeltaCue process is a pseudocolor raster image
that overlays on the original images. Non-zero pixels within this
overlay image represent change detections and are color-coded
according to material change type. If you open the image in the
regular IMAGINE viewer, under the Raster Options tab, select
Pseudo Color as the Display As option. These images are
automatically opened as pseudocolor images in the DeltaCue change
display viewer.
6
Process Inputs/Outputs
Change Detection Using DeltaCue
What is DeltaCue?
DeltaCue is a software package specifically designed to help you
identify changes of interest in remotely sensed imagery acquired on
two dates. The software provides a series of algorithms, procedures
and automated processing steps central to change detection in a
user-friendly form that helps efficiently manage the image
processing activities associated with this task.
The DeltaCue change detection process operates on coregistered image data and performs a differencing operation as
its core change detection process. Whenever you difference two
images, all kinds of change are present in the output. A key
feature of DeltaCue software is its approach to finding changes
of interest to you. It first tries to distinguish significant change
from insignificant change and then helps you identify changes
that are of specific interest to your application.
Significant change is any real change in the landcover materials
present in the two dates of imagery. For example, a new road or
building or a change in soil conditions are considered significant
change, though they may not be of interest to you. Insignificant
change is change due to sensor noise, atmospheric differences, or
image misregistration. These apparent changes are not due to scene
feature changes and are never of interest.
DeltaCue software allows you to eliminate insignificant change in a
number of ways. The process provides a threshold mechanism that
eliminates change based on change magnitude. Small-magnitude
change due to sensor noise or random reflections may be eliminated
by applying a minimum threshold for significant change. Also, a
maximum threshold is available to eliminate very large apparent
changes, such as those due to presence/absence of clouds or cloud
shadows. A misregistration filter is provided to help eliminate
apparent change due to pixel misregistration between the two dates
of imagery.
Once you have thresholded the change output, the remaining
changes evident in the output are considered significant change. In
general, however, this output will represent a mixture of different
types of change due to various effects. For example, a new road
might be included alongside changes in agricultural fields. The
DeltaCue software provides several mechanisms for distinguishing
changes of interest from other changes.
What is DeltaCue?
7
One mechanism for identifying changes of interest is the change
detection algorithm used. DeltaCue change detection algorithms first
apply a transform to the image data before performing the
differencing operation. These transforms are intended to better
enhance those sorts of landcover materials you are interested in. For
example, if you are interested in vegetation changes, the Tasseled
Cap greenness transform is a good indicator for vegetation, and
hence changes in vegetation are highlighted in the difference of the
greenness between dates. The primary color algorithms are best
suited for detecting man-made objects that appear red, green, or
blue. Thus, the choice of change detection algorithm eliminates
changes that are not of interest to you.
The second mechanism for identifying changes of interest using
DeltaCue software involves change filters. Change filters eliminate
uninteresting change based on spectral and spatial characteristics of
the detected significant change areas. You can eliminate change
based on the spectral characteristics of the landcover materials in
time 1 or time 2 and you can restrict the output to specific transitions
between materials based on spectral properties. For example, you
can eliminate change involving vegetation in time 1. You can also
eliminate changes that do not involve a change from one type
material to another. If you are only interested in changes from
vegetation to pavement, the software provides mechanisms for
filtering out all other changes, leaving only the change of interest.
You can also filter change based on spatial properties. Following the
change detection difference operation, regions of change are
identified and these regions loosely correspond to objects within the
scene. The spatial characteristics of these regions, such as area or
elongation, can be used to distinguish a change of interest from
other changes. DeltaCue software provides filtering on several
geometric properties that are useful in different applications. The
combination of spectral and spatial filtering allows you to weed out
changes that are not of interest, leaving only changes that meet your
interest criteria.
DeltaCue is designed to address the needs of both advanced users,
such as Image Scientists, willing to experiment with the process, and
production users, such as Image Analysts, who are primarily focused
on deriving a product. Typically an Image Scientist will develop a
change detection methodology for a given task and then provide
those procedures to the production analysts to perform on numerous
image pairs of similar characteristics. DeltaCue provides a series of
algorithms, filters and automated processing steps that the
advanced user can easily test and adjust to achieve optimized results
for a given change detection scenario (image type, environment,
change of interest, etc.). The procedures and parameter settings
used can be saved in a parameter file for future use by a production
analyst on other image pairs under a similar scenario.
8
What is DeltaCue?
These processing parameters can be subsequently implemented by
a production user with very little input other than input file names,
parameter file name and output file name. DeltaCue also
incorporates customized ERDAS IMAGINE viewers with many
features specific to the process of viewing and interpreting change
detection results. The DeltaCue change display viewer presents the
results of the change detection process in a viewer designed for
detecting change in a broad area search mode. An additional Site
Monitoring Mode Viewer provides tools for discrimination of subtle
image characteristic changes in a specific area of interest. Finally,
the software also provides image accounting procedures in order to
help you manage the resulting ancillary image files of successive
iterations and documents the various procedural steps implemented
in a session file.
Change Detection
Basics
Change detection within the context of digital image processing of
remotely sensed imagery of the earth’s features can be approached
from broad area search or site monitoring perspectives. In the first
case you wish to efficiently and accurately identify changes of a
particular type across a large area, potentially encompassing many
image pairs to cover the area spatially. For example you might be
interested in new roads across an entire country between two dates.
In the site monitoring scenario you are investigating a spatially
limited area, but perhaps has many images of that location over a
period of time. In this case the detection of a new road does not
require much more than visual inspection of the images; however
the detection of more subtle changes, which could aid in the
interpretation of less obvious changes, may require more advanced
processing and visualization techniques. The former case lends itself
well to the automation of procedures while the latter requires more
human interpretation.
Change detection entails different procedures and techniques, some
of which are standardized, while many depend upon the given
application. For example, in order to compare one image to another,
a pixel in one image must correspond to the same ground area in the
accompanying image. Image-to-image registration is a standardized
procedure that applies to any image pair. Conversely, filtering out
detected change based on a particular size is dependent upon the
size of the change phenomena in which you are interested. This will
vary from user to user, session to session.
Change Detection Basics
9
While many things may have changed in the imagery between the
two dates, normally you are only interested in a limited number of
specific types of change. Incidental changes confound the detection
of these changes of interest. Illumination changes related to image
acquisition at different times of the day or different times of the year
may result in an unchanged object appearing to have changed in its
brightness values. Obviously, deciduous trees will appear quite
different in their leaf-on state than their leaf-off condition. River and
tide levels may fluctuate from time 1 to time 2, snow and ice may
cover a landscape, vehicle traffic will likely move and agricultural
lands will cycle through fallow periods and different crop types. For
some users these changes are of interest, while for others these
image changes represent clutter to be filtered out of their analyses.
Filtering through the myriad of changes between two images is
achieved through a combination of the selection of change detection
algorithms appropriate for a given phenomenon and the
implementation of post-processing steps designed to eliminate
detected change not fitting criteria indicative of a change of interest.
Preliminary Steps
Pre-processing of an image pair consists of standardized procedures
that help eliminate sources of spurious or invalid change. Image
registration and radiometric normalization are two pre-processing
steps that are common to nearly all change detection studies and
therefore can be standardized and automated to a degree. An
additional technique to speed processing and improve the
appearance of the results includes cropping the images down to the
common area in the image pair.
Co-Registration
In order to be certain that the pixel value at row i column j in the
image at time 1 is being compared to the same ground area
represented by the pixel value at the same row/column location in
the image at time 2, the two images must be precisely registered one
to the other. In imaging terminology the two images should be coregistered. This process guarantees that a given pixel in one image
will correspond to the same pixel in the other image. This does not
imply any correction for sensor acquisition distortions, terrain
distortions, or other geometric transformations of the pixel grid.
Therefore the pixels in the co-registered images may not be square
and they could contain geometric distortions; however each pixel, no
matter how distorted, corresponds in a one-to-one relationship with
a pixel in the other image. Figure 1 shows the co-registration of one
image (blue) to a reference image (green), maintaining the
geometric distortions of the reference image. The image to be
registered is sometimes referred to as the slave image and the
image to which the slave must be registered is called the master (or
reference) image.
10
Change Detection Basics
Figure 1: Co-registration of a slave image (blue) to the
master image (green).
In order to perform the transformation of the slave pixel grid to the
master image a resampling of the slave image must occur, producing
a new, co-registered image. This resampling process alters the
values of the pixels, typically through an interpolation process, and
could therefore have an impact on the results of any change
detection process. For this reason it is best to register the images
using the fewest resampling procedures possible.
A co-registered image pair is not necessarily geo-referenced,
meaning that the image grids are aligned to a known, earth-based
spatial framework or coordinate reference system, such as Universal
Transverse Mercatur (UTM) or latitude/longitude coordinates.
Radiometric Normalization
Scene illumination differences and radiometric distortions in the
sensor can result in unchanged features having different brightness
values in one image relative to another image. This will result in
apparent change detected where none has actually occurred. This
effect can be minimized by ensuring that the images are collected
under similar illumination conditions (images of the same time of day
and time of year) and through the avoidance of images impacted by
either atmospheric haze or sensor anomalies.
Change Detection Basics
11
To best eliminate or greatly minimize the effects of scene-to-scene
radiometric difference, DeltaCue software performs a radiometric
normalization of the two images. A radiometric correction to
reflectance is also a possible solution, although this process typically
requires additional information, as well as more processing time, and
must be performed outside of DeltaCue. Additionally, studies have
shown that in the context of change detection, the results achieved
by simply normalizing one image to another are as good, and in
some cases better, than attempts at radiometrically correcting both
images to reflectance and then performing change detection (Collins
and Woodcock, 1996; Chavez and MacKinnon, 1994; Jensen et al,
1995; Varjo, 1996; Olsson, 1995; Hall et al, 1991).
Change Detection
Methods
Change Detection Algorithms
There are numerous algorithms and procedures for detecting change
between two images. They can be broadly divided into two
categories: transformational techniques that produce a change
image for which a change/no change threshold must be established,
and change classification techniques where image pixels deemed to
represent change are directly detected in the source imagery and
mapped. DeltaCue focuses on the former set of techniques.
The most direct technique is to subtract one image pixel from the
corresponding pixel in the other image. Theoretically, if no change
has occurred, the difference between the two is a zero mean noise
distribution. Assuming the subtraction of time 1 from time 2, pixels
that have gotten brighter would have positive values and pixels that
have gotten darker would have negative values. The distance of that
value from zero indicates the degree of change which has occurred.
The resulting change image is a single band, gray scale image for
each band to band comparison in the original image pair.
DeltaCue software uses a symmetric relative difference formula to
measure change, as in:
T2 – T1 T2 – T1
------------------ + -----------------T1
T2
Dividing the difference by the pixel’s value at time 1 and time 2
allows the derivation of a change image that measures the
percentage change in the pixel, regardless of which image is chosen
to be the initial image. A pixel that had a value of 20 at time 1 and
a value of 80 at time 2 would have an absolute difference of 60, and
a percentage change value in the change image of 375%:
[(80 – 20) / 20 + (80-20)/80] * 100 = 375%
Another pixel with a value of 140 at time 1 and 200 at time 2 would
also have an absolute difference of 60, but its percentage change
would only be 72.86%:
[(200 – 140 ) / 140 + (200-140)/200] * 100 = 72.86%
12
Change Detection Basics
In most cases it can be assumed that the percentage change of a
pixel’s brightness value is more indicative of actual change in the
image than simply the absolute difference.
A percentage difference change image can be created by simply
differencing every band in one multispectral image from its
corresponding band in the accompanying image, although this will
result in as many change images to interpret as bands in the image.
There are many different ways to condense the information
contained in the bands of one image into one which can then be
differenced with the other image’s condensed band image. For
example, one can difference each pixel’s overall magnitude across all
bands.
In addition to using techniques to compress the information
contained in multiple bands, other processing techniques can be
used to highlight certain spectral features and suppress others. If a
phenomenon of interest is known to exhibit certain spectral
behaviors in multi-temporal imagery, then procedures can be
implemented to exploit that behavior. For example, recent forest
clear-cuts are associated with a decrease in near-infrared bands
(Landsat Thematic Mapper band 4) and an increase in shortwaveinfrared bands (TM band 5). In order to highlight this particular
spectral feature, a differencing of the ratio of bands 5:4 from time 1
to time 2 is an effective method of detecting this change.
Image transformation procedures based on scene phenomenology,
such as the Tasseled Cap Transformation (Kauth and Thomas, 1976;
Crist and Cicone, 1984; Crist, 1985; Collins and Woodcock, 1996),
are another option. This procedure converts imagery from the
original bands to a set of components that correspond to scene
elements such as soil brightness, greenness, haze and others. The
resulting phenomenological components can be differenced and a
change image of soil brightness or greenness can be created. There
are nearly as many ways to draw out change from imagery as there
are ways to process the data; the key is to use a method which
highlights the change of interest, if it is known.
Change Threshold
A deviation from zero in the change image may not be the result of
actual change in scene elements but could be due to the natural
spectral variability of objects in the images. Even when accounting
for illumination differences and other exogenous, non-change
related effects, rarely will an unchanged target have exactly the
same value in the pixels that represent it in the two images.
Therefore, it is necessary to establish the distance from zero at which
an apparent image change is representative of a real physical change
in landcover.
Change Detection Basics
13
The setting of this change threshold can be done arbitrarily by
selecting a percentage change value or you may decide upon a
statistical value, such as a number of standard deviations away from
the mean. In a change image the mean is typically centered at or
near zero (assuming the majority of the scene is not change). The
data are characteristically distributed as a Gaussian (or Gaussianlike) curve, the tails of which represent the negative and positive
changes. Figure 2 Shows the histogram of a change image with data
mean centered at or near zero (no change) and distribution tailing
off to increasing levels of positive and negative change.
Figure 2: Histogram of a change image.
Negative
Change
Positive
Change
Figure 3 shows the Histogram of a change image illustrating change
thresholds established through the use of image statistics, 2.5
standard deviations from the mean. Data falling beyond these
thresholds are classed as change.
14
Change Detection Basics
Figure 3: Histogram of a change image with thresholds
2.5
Negative
Change
Threshold
σ
Positive
Change
Threshold
Establishing a change threshold is a matter of deciding which points
along the tails are the points where the normal variability of the
image ends and values corresponding to actual scene changes
(either negative or positive) begins. The use of a number of standard
deviations normally implies that the threshold for negative and
positive changes are equal, only differing in their signs, for example,
+/- 2.5 standard deviations (Figure 3). However, there is no reason
to assume that the negative and positive change thresholds are at
equal distances from the mean; that depends on the vagaries of each
image pair and the change detection algorithm used.
Change thresholds can also be set based on the percentage change
that was detected in the percentage difference image. If a threshold
of 30% is established, then a pixel whose value in the image
increased by 30 % or more would be mapped as change:
[(200 – 170) / 170 + (200-170)/200] * 100 = 32.64%,
32.64% > 30%, therefore is mapped as change.
Real change exists above this threshold, or below if it is a negative
threshold. Similarly an outer threshold could be set to mask out
extreme changes, such as percentage changes over 300%.
Change Detection Basics
15
Figure 4 shows the histogram of a change image with upper and
lower bounding thresholds for both negative and positive change.
Introducing upper and lower bounds to the change regions of the
histogram potentially introduces additional resolving power on the
visualization and also introduces an additional level of complexity.
The outer bounds can be useful when changes that are not of interest
are found in the extreme tails of the change image, such as with
clouds or cloud shadows.
Figure 4: Histogram of a change image with bounding
thresholds.
Negative
Change
Positive
Change
Upper
Bound
Lower
Bound
Filtering Unwanted
Change
In the case where a user has no particular phenomena of interest in
mind and wants to see any and all change between two dates of
imagery, you may begin to analyze the change image immediately
after the thresholding process. However in many instances, you will
have some idea as to what kinds of changes are important and which
are trivial. You need tools to winnow out detected changes that are
not of interest. These are not false change detections, merely
detections not pertinent to the current analysis.
Spatial Filtering
For instance, a user interested in new buildings of a given size will
not care about change detections that are much smaller than a
typical building of interest. In an image with 2-meter pixel
resolution, single pixel change detections can most likely be
considered spurious and removed.
16
Change Detection Basics
In that same scenario, you may decide to disregard contiguous
groupings, or “blobs” of detected change below the size of a
building. However, if the change detection method or
thresholding process results in only a partial detection of a
building, then a filter that only looks for groupings as large as a
building, would result in partial building detections being
omitted. Therefore it is advisable to allow for detection sizes at
least one half the size of the change phenomenon of interest.
It is also possible to filter change regions in terms of additional
characteristics besides area. DeltaCue software provides the ability
to filter on several types of geometric characteristics, including the
length of the principal axes, geometric compactness, and elongation.
The change region blob may be thought of as a mass in twodimensional pixel space. The center of mass or centroid of the blob
defines the mathematical center point for the region. The centroid is
one of several geometric moments (analogous to moments of inertia
in physics). The major principal axis is a line through the centroid
along the bulk of the region’s mass. The minor principal axis is
perpendicular to the major axis.
Figure 5: Change Region Blob
Change Blob
Minor Axis
Centroid
Major Axis
The length of these axes essentially defines the characteristic
dimensions of the region, particularly for regularly shaped regions.
These lengths can be used as shape discriminators. For example, to
detect new buildings of a particular size, you could filter based on the
length of the major axis using the length of the longer side of the
building. A major benefit of using principal axes is that you do not
need to know the orientation of the building in order to filter the
change. For regular shapes like buildings, the major axis will
correspond to the major dimension of the building.
Elongation is defined as the ratio of the major principal axis length
to the minor axis length. For regularly shaped change regions,
elongation is a measure of how extended the region is. A circular
region or square would have an elongation of 1.0 whereas a long thin
region like a new road would have an elongation value that is much
higher.
Geometric compactness is another shape characteristic that can be
used to filter change. This quantity is defined as the area of the
change blob divided by the product of the major and minor axes:
Change Detection Basics
17
A
C = ----------P1 P2
where A is the region area, P1 is the major axis and P2 is the minor
axis of that region. As the change region grows longer in one
dimension, the compactness tends to decrease. A perfectly square
area would have a compactness of one. The table below shows
compactness values for various rectangles and shows how
compactness can be used to discriminate various shapes.
Table 1: Comparison of Compactness Values for Various
Rectangular Shapes
Ratio of Rectangle
Length to Width
Compactness
1 (square)
0.785
2
0.698
3
0.589
4
0.502
5
0.436
10
0.260
15
0.184
20
0.142
Further information on setting the various parameters of the spatial
filter is available in "Change Filtering" on page 30.
Misregistration Filtering
Another type of spatial filtering that can be applied to the
thresholded change image is a procedure to adjust for known levels
of misregistration between the two images. Images that have not
been precisely co-registered will exhibit certain predictable
behaviors in the resulting change image. Along roads and the edges
of buildings there may be detections of change related to the slight
mismatch of the pixels being compared. A misregistration of one
pixel across an image pair will result, for example, in a pixel
representing the edge of a road being compared to a pixel
corresponding to the vegetated field on one side of the road. At the
same time on the other side of this linear feature the pixel
corresponding to the water body on the other side of the road may
be compared to a road pixel in the other image.
Figure 6 shows a schematic representation of properly and
improperly registered image pairs: A represents a properly coregistered pair while B represents a mis-registered pair.
18
Change Detection Basics
Figure 6: Properly and improperly registered image pairs.
Typically a set of pixels with one sign of change (negative or positive
depending on the targets involved) will occur on one side of the
linear feature and another set of pixels with the opposite type of
change will be found on the other side of the same feature.
Figure 7 shows a schematic representation of the effect of
misregistration on the change image: Red and blue colors indicate
erroneous detected change. This problem can be minimized through
careful registration, but frequently some level of local
misregistration is unavoidable.
Figure 7: Representation of the effect of misregistration.
One way to compensate for this effect is to pass a filter over the
change image that will investigate incidences of detected change and
ascertain whether or not that change may be due to a
misregistration effect. For every pixel highlighted as change, the
algorithm can go back to the original images and see if, for that pixel,
there exist in a search window of a given size (3x3, 5x5, etc.) two
spatially mismatched pixels that, had the image been shifted over a
slight amount, there would have been no change detected.
Change Detection Basics
19
Figure 8 shows how misregistration has caused pixel A to be shifted
to the location of pixel A’ in the time 2 image, and a change to be
detected between the value of A and B’. A search in the 5 x 5 window
around B’ reveals a pixel, A’, which matches pixel A in time 1. This
misregistration is confirmed by checking B’ against B. If not only A
matches A’ but also B’ matches B then the change from A to B’ is due
to misregistration. The shift from B’ to B is with the same amount of
shift from A to A’ but with opposite direction.
Figure 8: Detection of Misregistration.
Spectral Filtering
Aside from filtering detected change based on its spatial
characteristics, change pixels can also be filtered on the basis of their
spectral nature. Spectral characteristics can be associated with
physical landcover types and hence, spectral filtering is a means of
filtering out unwanted change based on landcover.
DeltaCue software uses a form of spectral segmentation which
classifies change pixels in the Time 1 and Time 2 images using
unsupervised classification. These classes represent the before and
after landcover classes. The combination of the two classes
represents transitions from Time 1 (before) landcover to Time 2
(after) landcover.
20
Change Detection Basics
Figure 9: Spectral segmentation using unsupervised
classification.
Before Classes
After Classes
Transitions
The DeltaCue change display viewer allows you to filter out
unwanted change based on either the before class (material), the
after class (material), or the transition from one class to a second.
For example, if you are only interested in changes in which
vegetation became bare soil, you can filter out all before classes that
are not vegetation and all after classes that are not bare soil. The
remaining transitions represent changes from vegetation to bare
soil. You can get even more specific by filtering out specific
transitions, for example, trees to bare soil. The remaining change
pixels represent changes from spectrally similar before and after
materials.
The spectral filtering process in DeltaCue does not assign
information to the spectral classes; they are simply numbered
classes. It is your interactive interpretation that allows you to
remove or keep spectral change classes on the basis of their
interpreted information class. The “DeltaCue Tutorial” on page 35
provides an example of how you could use the spectral segmentation
tool to remove change classes not of interest on the basis of spectral
classes.
Material Filtering
Spectral segmentation is an effective method for separating out
change pixels into different data classes, but it does not tell you what
those classes correspond to in terms of scene features. You must
interpret this information. Without rigorous image calibration and
atmospheric corrections combined with spectral signature libraries
and ground truth data sets, it is very difficult to identify information
classes such as landcover types or scene features.
Change Detection Basics
21
DeltaCue makes use of the Tasseled Cap Transformation procedure
to conservatively filter out change pixels on the basis of their
material type in either the time 1 or time 2 image. This technique is
intended to assist in further removing changes not of interest, such
as phonological changes (leaf on/leaf off comparisons) or waterrelated changes.
Analyzing Change
After the filtering of a change image, you are left with an image
where pixels that have been classed as change are coded, perhaps
for the direction of the change (bright to dark-negative or dark to
bright-positive), the intensity or magnitude of the change (as a
percentage change or as classed range of change) or as change of a
particular type (soil brightness change, greenness change, etc.).
Areas of no change are classed as zeros. While this result could be
directly exported into a GIS as a model input or for decision support,
the majority of the time the results are overlain on the original
imagery for you to evaluate. Ultimately, a human interpreter looks
at the highlighted pixels in the original imagery and makes a decision
as to the nature of the change, the importance of that change, the
need for further investigation or a decision on a course of action.
This process of interpretation typically consists of overlaying the
change on the original imagery and then switching between the first
and second date of imagery. Using the graphical user interface of an
image processing software package the analyst makes use of the
existing tools to assist in their interpretation. In the case of ERDAS
IMAGINE software this includes the ability to swipe or flicker between
two dates of imagery, to change the band combinations of the
imagery, to alter the opacity or color of a given change class and
many other functionalities.
The highlighted pixels of the change image act as a “cueing tool” to
guide the eye of the interpreter to specific regions of an image for
analysis. The ultimate product of the exercise may be a series of
point coordinates that you locate on the basis of his/her
interpretation of the change results. Filtering out the changes not of
interest, you would only flag those highlighted change areas that
have been evaluated as truly of interest.
As an example, let us consider a user interested in the construction
of new structures at industrial facilities on the outskirts of an urban
area. Using the appropriate change detection algorithm and filtering
procedures, the detected change could be filtered of change due to
misregistration, small change areas associated with vehicles, and
vegetation changes of similar size to the buildings of interest. You
would be left with a series of change pixels that fit the spectral and
spatial description of the target of interest. However among the new
industrial structures there would be some new residential structures
also identified. Having a similar spectral composition and spatial size
these features would fit the same description as the new industrial
structures. On the basis of his/her interpretation (based on the
interpreted context of the structures) you could flag just those points
that have been interpreted to be the change of interest and leave out
the other valid, but not-of-interest, changes.
22
Change Detection Basics
For some applications there may be a desire to use the imagery to
not only answer the question of the existence of change and
determine its location, but to also quantify that change. For
watershed modeling purposes the existence and point location of a
forest clear-cut or impervious surface may not be enough. There
may be a desire to delineate polygons of forest loss for input into
models or acreage estimates of impervious surface growth. In these
cases you would do more than define a point where the change
occurred but select out the changed block of pixels for inclusion into
a GIS layer for model integration or statistics generation.
DeltaCue Change
Detection
DeltaCue Change Detection software brings together a collection of
image processing techniques and assembles them into an integrated
procedure. While principally designed to address the needs of broad
area search scenarios, DeltaCue also includes tools for site
monitoring applications. Ideally a fully automated process for
detecting change is desired; however the variety of possible data
types, image environments, targets/changes of interest and image
acquisition parameters leads to a complexity that renders
automation nearly impossible. DeltaCue attempts to bridge the gap
between the highly skilled Image Scientist (IS) who can develop a
sophisticated change detection methodology for a specific
application and the Image Analyst (IA) who is responsible for the
processing of many images in short turn around time.
DeltaCue achieves this by linking together a series of procedures
best suited for a range of potential change detection applications in
general. The software allows the IS to easily test out settings and
algorithms to achieve best results. Through a process of iteration
and testing using customized viewer tools, you establish a useful set
of procedures for their particular change detection task. These
procedures are saved as a parameter settings file which can then be
used to automatically select the appropriate algorithms and filter
settings in subsequent processing by an IA. The interpretation and
evaluation of results are done similarly by both ISs and IAs.
Processing Strategy
DeltaCue Change Detection
The change detection strategy used by DeltaCue in broad area
change searches, is to allow as much change as possible to pass
through the detection process and then to limit the amount of
detected changes with subsequent runs of various filters. It is critical
that all change is detected and initially passed to the filters. While it
is desired to minimize the amount of time spent evaluating
highlighted change that is not of interest, this should not be done by
restricting detections to the point of missing valid changes of
interest. Excessively conservative thresholds may prevent these
changes from being passed through to the viewer and you will not be
cued to that area of the image. Therefore, it is advisable to allow
generously open thresholds that will pass more change than may
initially be desired with the idea that subsequent filters will assist in
winnowing out the change pixels not of interest.
23
If a user has no particular type of target in mind, but is interested
generally in all things that have changed between the two dates,
then an algorithm such as the magnitude difference would be
appropriate. All changes, regardless of associated phenomena, are
detected on the basis of the overall magnitude of the pixel’s
brightness change. If on the other hand you know that they are only
interested in changes related to soil disturbances and not
vegetation-related changes, then a phenomenological-based
approach, such as Tasseled Cap Differencing, would help to initially
filter out some of the intense vegetation changes that would
otherwise be passed by a magnitude difference approach. This
should allow you to set a more open threshold on the soil brightness
difference image, without fear of overwhelming the change image
with extreme vegetation-related, changes that while valid, are not of
interest.
Pre-Processing
DeltaCue has been designed to automate as much as possible the
repetitive and time consuming steps of pre-processing. Tasks such
as cropping an image pair to a common coverage area or performing
image to image normalization are for the most part standardized
procedures.
Co-Registration
The task of automated image to image co-registration has been a
challenge for the remote sensing community for many years. As
DeltaCue is an integrated add-on module to ERDAS IMAGINE, it
integrates with the available tools in that software for this process.
AutoSync is an automated image registration tool in ERDAS IMAGINE
used to automatically generate control points and co-register the
image pair. Refer to the AutoSync documentation for more
information.
DeltaCue assumes that you bring to the process two acceptably
co-registered images. The images do not need to cover exactly
the same area, as the automated cropping procedure will ensure
this, but they should be co-registered to within a half of a pixel
as a rule of thumb. The exact level of accuracy will depend upon
the data source and your needs. Also if certain areas within an
image are of more interest than others, a lower overall accuracy
level, as measured by the Root Mean Square (RMS) error of an
image transformation, is not that bad if the areas of interest are
well registered.
24
DeltaCue Change Detection
Radiometric Normalization
There are a number of approaches to image-to-image normalization.
DeltaCue implements an automated procedure that forces the mean
and standard deviation of the time 2 image to match the same
statistics of the time 1 image. This is done through the calculation of
two band-wise linear transformation coefficients from the mean and
the standard deviation spectra of the two images. This linear
transformation is then applied to the time 2 image to force its
statistics to match those of the time 1 image on a band to band basis.
This procedure is based on the following assumptions:
environmental effects such as illumination or homogeneous haze are
the dominant source of differences between the two images and
there exists a band-wise linear transformation between pixel values
of the two images. Violations of these assumptions, such as clouds,
can cause a less than optimal normalization which could impact the
change detection results. In order to address this problem, DeltaCue
allows you to specify the presence of clouds in either of the images.
The normalization procedure will then implement routines to remove
the effect of the clouds and cloud shadows on the overall statistics of
the images.
NOTE: This will not remove change detections corresponding to the
presence of clouds or cloud shadows. This procedure will simply
minimize the effect of these clouds on the normalization routine.
Non-linearities in sensor response and differential haze across an
image will also affect the quality of the image normalization.
Should you have images that have already been normalized or even
corrected to reflectance through an external procedure, there is the
possibility of opting out of the normalization procedure and running
the images as is. If there is doubt as to whether the images should
be normalized, it is recommended that you run the normalization
routine in DeltaCue.
Change Algorithms
DeltaCue incorporates a set of change detection algorithms which
are described below. Future versions of the software could include
new algorithms or alterations of the current set. The current
algorithms are: Magnitude Differencing, Tasseled Cap Soil
Brightness and Greenness Differencing, Primary Color Differencing,
Single-Band Difference, and Band-Slope Difference. You can run just
one or iteratively experiment with them all. All of the change
algorithms use the following formula for computing relative
difference:
T2 – T1 T2 – T1
------------------ + -----------------T1
T2
Magnitude Difference
As the name implies the Magnitude Difference algorithm calculates
for each pixel its brightness magnitude across all bands in the image
based on the following formula:
DeltaCue Change Detection
25
n
Mi =
2
∑ BVij
j–1
A pixel’s magnitude value at time 1 is then subtracted from its value
at time 2 and the relative difference is computed using the formula
above.
This provides a measure of change across all the bands in an image.
This detection approach works well for many types of phenomena
that change a pixel’s brightness value in all spectral bands. This
would include presence/absence events when the background is
dissimilar to the target, for instance when a bright vehicle moves off
an asphalt parking lot. Likewise standing water changing to bright
dry sand as river levels recede would be another phenomenon that
lends itself well to detection by the magnitude differencing process.
Tasseled Cap Differences
The Tasseled Cap Transformation involves the application of linear
transformation equations to the original data sets based on
empirically derived transformation coefficients specific to the sensor.
The new transformed image components correspond to scene
phenomena such as soil brightness and greenness. These
components are then differenced to provide a change image of the
given component, that is, positive or negative changes in soil
brightness or greenness.
DeltaCue specifies either a soil brightness or greenness difference
image as a change detection algorithm; however it will compute the
entire transformation the first time either of the two methods is
used. Subsequently, should you choose to use the other Tasseled
Cap method, DeltaCue will not compute the transformation again but
use the pre-existing components, resulting in a much faster
difference image calculation for the second method called.
The Tasseled Cap difference images produce change images which
can be very different from the magnitude difference image. The
greenness difference image will typically provide more detailed
discrimination of subtle vegetation changes. Extreme changes, such
as forest removal, building construction and fire scars will at times
be detected in the Tasseled Cap procedures regardless of whether
they are related to the phenomenon of the component in question.
These same extreme change events will also be present on the
magnitude and principal component change images as well.
Finally, the Tasseled Cap Transformation requires multispectral
imagery, which means it cannot be used with panchromatic imagery.
Not all multispectral sensors have had Tasseled Cap Coefficients
calculated for them. In those cases, routines based on this
transformation will not be available. The transformation can also be
sensitive to the atmospheric conditions of the original imagery and
sensor response anomalies and noise.
26
DeltaCue Change Detection
Primary Color Differences
DeltaCue software provides a change detection algorithm that
highlights changes in scene objects that are predominated by one
primary color. The algorithm first thresholds pixels in both the time
1 and time 2 images based on how red, blue or green an object
appears and then those pixels that are considered to be of one
particular color are differenced as in a normal image differencing
process. The color of a pixel is defined using a spectral angle
transform. Consider a three-dimensional color space in which the
band numbers are ordered in wavelength ascending order. The pixel
spectrum is represented by
r
P = ( P1 , P2 , P3 )
The cosine of the spectral angle transform of the pixel vector with a
reference vector
r
R = ( R1 , R2 , R3 ) is given by
N




R
P
∑
i i


i =1
cos α =
1/ 2
1/ 2 

  ∑ R i2   ∑ Pi2  
  i
 
  i
Reference spectra representing the red, green, and blue primary
colors are given by
RED = (0,0,1)
GREEN = (0,1,0)
BLUE = (1,0,0)
Using the red, green, and blue color vectors as references gives the
following red, green, and blue transforms:
DeltaCue Change Detection
27




P3


cos(α RED ) =
1/ 2 

  ∑ Pi2  
  i
 




P2


cos(α GREEN ) =
1/ 2 

  ∑ Pi2  
  i
 




P1


cos(α BLUE ) =
1/ 2 

  ∑ Pi2  
  i
 
The primary color change detection algorithm applies a given
transform to each image and then differences the result using the
regular DeltaCue difference spatial model. Since false change
indications can be obtained from spectral angles far away from the
primary color reference, the transformed cosine value is first
thresholded before the difference is taken. This prevents changes in
greenness, for example, from showing up as a change in redness
even though the cosine of the spectral angle changed. If neither
angle was close to red to begin with, then neither is included in the
final result. Essentially, the color threshold separates things of one
given color from other scene elements and then the change
threshold will define changes within those pixels.
Detecting primary color changes in a high spatial resolution image,
such as a multispectral QuickBird image, can be quite useful for
tracking changes in the location of differently colored vehicles or in
assessing changes to buildings and other painted structures. This is
because an entire pixel covers a surface painted a given color. In the
case of medium resolution imagery, such as Landsat TM or SPOT
data, each individual pixel will most likely cover a variety of different
colored materials. Unless an entire 30 meter by 30 meter area is
covered by a given color, the chances are that it will not pass the
initial color threshold and it will not be detected as being of that
color.
28
DeltaCue Change Detection
The primary color difference algorithm assumes that the input
imagery is from a multispectral scanner that has a blue band (~0.4
– 0.5 µm) as band 1, a green band (~0.5 – 0.6 µm) as band 2 and
a red band (~0.6 – 0.7 µm) as band 3. This assumption is correct
for IKONOS, QuickBird or Landsat TM imagery, however for SPOT-5
imagery the behavior of the software is different. In SPOT-5 imagery
band 1 is green, band 2 is red and band 3 is a near-infrared band.
Therefore, running the blue color algorithm will actually highlight
changes in green objects. Similarly the red algorithm will result in
changes in objects dominated by near-infrared reflectance. You
should be aware of this feature of the software, as it could lead to
confusing results in multispectral imagery where the bands are not
in blue-green-red order. Also, by reordering the layers in an image
(using the Layer Stack functionality in IMAGINE), an advanced user
could apply the color change detection algorithm to non-primary
color bands, such as near and mid-infrared bands. The primary color
change detection algorithm is disabled when single band or
panchromatic imagery are used as inputs.
Single-Band Differences
Single-band differences are sometimes useful if a particular change
phenomenon occurs primarily in one band or if only single-band
imagery is available. DeltaCue provides the capability to produce a
change result based on a single band relative difference. While
simple and intuitive, this single-band difference can result in a large
amount of insignificant change being detected. This technique is
most appropriate for use with panchromatic imagery or when used
in conjunction with other filters.
Band-Slope Differences
The difference in band slope between adjacent bands can also be a
useful indicator of change. DeltaCue provides a simple band-slope
change detection algorithm. The slope between adjacent bands j and
j+1 are computed, as in
T1 = B1 [ j + 1] − B1 [ j]
T2 = B 2 [ j + 1] − B 2 [ j]
The relative difference of these quantities is then computed using the
regular DeltaCue relative difference formula. The band-slope
computed is always specified by the lower band j, which ranges from
1 to N-1. N is the number of bands for each of the inputs used in the
algorithm for change detection.
DeltaCue Change Detection
29
Change Threshold
The setting of a change threshold is a critical step in the change
detection process. DeltaCue provides a flexible tool for setting this
threshold. A user could simply use a specified percentage change
value as a threshold, for example, any pixel whose brightness value
has changed by more than 90% is considered change. Unfortunately,
change images can be very different from one image pair to the next,
from one application scenario to the next. Knowing what the
percentage change value should be without first investigating the
change image is not always possible.
While many image processing packages provide the flexibility to use
different thresholds, DeltaCue uses a unique, interactive tool for
change threshold setting that allows you to immediately see the
effects of changing the threshold before creating the change image.
Using the change image’s histogram as the interface, you can move
the threshold using the mouse relative to the histogram. This will
change the masked change image which is simultaneously displayed
alongside the threshold tool. While this process does still require user
interpretation, it provides a more readily understandable means by
which to do so. For the Image Scientist this tool speeds up the
process of developing an optimized procedure for their change
detection task.
The DeltaCue process allows successive iterations to be tested,
saved and evaluated. You may decide to re-run the same change
detection algorithm using a different change threshold. Additionally,
the change threshold tool also allows for more complex threshold
setting than normal image processing procedures. You could set
upper and lower limits to the range of interest should this suit their
needs (see Figure 4). For advanced users this could help to detect
subtle features within a change image by masking out extreme
changes in the image which exist as outliers on the tails of the
change image histogram. Additionally, you could decide to threshold
out change in a given direction (for example, things that got darker)
if they know that their phenomenon of interest is in the opposite
direction (only pass positive changes, things that got brighter).
Change Filtering
The DeltaCue change detection strategy is to allow a large amount
of change to pass the threshold process and then rely on filtering to
limit the final detected change to those pixels that most closely
match your change of interest. The DeltaCue software incorporates
both spectral and spatial change filtering. If Tasseled Cap
coefficients are available for the images, changes may be filtered
based on before and after material types (vegetation, nonvegetation, and water/shadow). The software also classifies each
change pixel into spectrally similar before and after states. These
classes can be used to interactively eliminate change.
Geometric properties of the change regions identified can also be
used to filter change. Once the change quantity is thresholded,
change regions become evident in the output. A change region is a
contiguous blob of connected pixels. As long as any part of the region
is connected via one of its eight neighbors to any other region, it is
considered part of the same region. For example, the pixels shown
below are considered one shape since the two regions are connected
through a common neighbor.
30
DeltaCue Change Detection
Also, the shape and size of the change regions can depend on
the change thresholds used. Thus, the spatial filter should be
used as a general filter with loose tolerances to filter out classes
of change, not necessarily precise characteristics.
Also, it is important to keep in mind that discrete change features
that abut one another are represented in the change image as one
continuous blob. For instance a vehicle may be parked next to a bare
field in time 1. In the time 2 image, the vehicle is gone and the field
is now full of corn. Both the vehicle absence and the greening of the
cornfield could be detected and pass a threshold as change. If you
are interested in detecting vehicle movements, not agricultural
change, he or she might set a filter to remove large blobs, while
retaining smaller regions corresponding to the size of a vehicle. The
region representing the vehicle in this case would be filtered out
since it would be part of the large region corresponding to the field.
Continuing the above scenario, you could target the vehicle change
of interest by first selecting the appropriate change detection
algorithm, such as the Tassseled Cap Soil Difference algorithm which
would help minimize the effect of the vegetation-related change. If
a particular colored vehicle was of interest the primary color change
algorithm would also be an option for use. Secondly, the material
filter could be set to filter out change pixels that appear to be
vegetation in the time 2 image; this would further reduce the
contribution of the vegetation change to the final change image.
Keeping the filter slightly larger than the object of interest will help
avoid the effect of removing a specific sized vehicle being removed
because it abuts a larger group of change pixels (such as the corn
field). Finally, the spectral segmentation routine will allow you to
perceive the difference of the vehicle from the nearby vegetation
pixels and interactively turn off those classes that correspond to the
vegetation. What should be left as cues in the change image are
those change pixels that most closely resemble the changes of
interest.
DeltaCue Change Detection
31
There is also a special kind of spatial filter available in DeltaCue that
works on spatial misregistration in the image. As described in
"Filtering Unwanted Change" on page 16, errors in image-to-image
co-registration can result in spurious change detections. DeltaCue
attempts to model and correct for these detected changes. You can
select different search window sizes and the number of misregistered
neighbor pixels to test for in their original images. This procedure is
appropriate to use when there is known misregistration in the
imagery.
Change Results Viewing
The final, filtered results of the change detection process are
displayed in the DeltaCue change display viewer. This viewer is
similar to the existing viewers of ERDAS IMAGINE, while
incorporating new features specifically designed to aid the
interpretation of change results. In particular, you can filter out
unwanted change based on spectral classes of before materials, after
materials, and transitions between classes. Chapter 5 provides a
detailed description of the functionality of the change display viewer.
Process Management
The above series of steps and procedures for change detection
normally results in numerous image files and notes of algorithm and
filter settings. If a user should return to a change detection activity
after some time, it can frequently be difficult to sort through the
various images and ancillary files in order to ascertain exactly which
procedures had been run and what settings were used in a given
process. DeltaCue helps you organize and manage this information.
The principal process management tool that DeltaCue uses is the
Project file (.dqw file). This XML file is used to store the history of a
change detection session, including the input images, the preprocessing steps run and their resulting image files, the change
detection algorithms used, thresholds established and their image
files, and the filters implemented and their settings. The Project file
documents each iteration performed and, through the change
display viewer, allows you to call up the results from any of the
iterations for comparisons. A user can also open up a previously run
project and pick up processing where he or she left off.
Another aid for you is the settings file (.dqs file) which allows one
user to save their process settings (change algorithms and filters) in
a file so that another user can apply them to a different image pair
under a similar change scenario. This is designed to allow a more
advanced Image Scientist to develop an appropriate change
methodology that could be easily implemented by an Image Analyst
with less image processing knowledge.
The settings file allows for the capture of algorithm and filter settings
that have been optimized for detecting a given change phenomenon
while minimizing changes not of interest, such as detecting new
roads in an image containing many other kinds of changes. While the
settings can be saved and applied to the detection of the same
phenomenon in another image pair, there are some image pair
parameters that are unique to each image pair, and therefore each
DeltaCue session.
32
DeltaCue Change Detection
An Image Analyst can specify a settings file when initiating a
DeltaCue session through the Wizard interface to assist in initializing
the parameters of the shape filter, the material filter and the
selection of the appropriate change detection algorithm.
See “DeltaCue Wizard Interface” on page 53.
However, the settings file will not be able to tell the analyst whether
the current image pair should be cropped, whether it has clouds in
either the time 1 or time 2 images, or the settings for the
misregistration filter, should it need to be run.
Ideally, the change threshold should be one number that is set for
all image pairs, but in reality there are many cases where it is
necessary to alter the threshold for a given pair.
In general the settings file will help the Analyst with the more
complex filter settings for a given application. With a minimum of
training you will be able to define the additional processing steps
necessary for the particular image pair at hand and can then focus
your attention on interpreting the change results and not on how to
derive the results.
References
Collins, J., and C. Woodcock, "An Assessment of Several Linear
Change Detection Techniques for Mapping Forest Mortality
Using Multitemporal Landsat TM Data," Remote Sensing of
Environment, vol. 56, 1996, pp. 66-77.
Chavez, P., and D. MacKinnon, "Automatic Detection of Vegetation
Changes in the Southwestern United States Using Remotely
Sensed Images," Photogrammetric Engineering and Remote
Sensing, vol. 60, no. 5, May 1994, pp. 571-583.
Jensen, J., K. Rutchey, M. Koch, and S. Narumalani, "Inland Wetland
Change Detection in the Everglades Water Conservation Area
2A Using a Time Series of Normalized Remotely Sensed Data,"
Photogrammetric Engineering and Remote Sensing, vol. 61,
no. 2, February 1995, pp. 199-209.
Varjo, J., “Controlling Continuously Updated Forest Data by Satellite
Remote Sensing,” International Journal of Remote Sensing,
vol. 17, no. 1, 1996, pp. 43-67.
Olsson, H., "Reflectance Calibration of Thematic Mapper Data for
Forest Change Detection," International Journal of Remote
Sensing, vol. 16, no. 1, 1995, pp. 81-96.
Hall, F., D. Strebel, J. Nickeson, and S. Goetz, “Radiometric
Rectification: Toward a Common Radiometric Response Among
Multidate, Multisensor Images,” Remote Sensing of
Environment, vol. 35, 1991, pp. 11-27.
References
33
Kauth, R., and G. Thomas, “The Tasseled Cap - A Graphic Description
of the Spectral-Temporal Development of Agricultural Crops as
Seen by Landsat,” Proceedings of the Symposium on Machine
Processing of Remotely Sensed Data, LARS, Purdue University,
West Lafayette, IN, 1976, pp. 4b41-4b51.
Crist, E., and R. Cicone, “A Physically-Based Transformation of
Thematic Mapper Data - The TM Tasseled Cap,” IEEE
Transactions on Geoscience and Remote Sensing, vol. 22, no.3,
1984.
Crist, E., “A TM Tasseled Cap Equivalent Transformation for
Reflectance Factor Data,” Remote Sensing of Environment, vol.
17, 1985, pp. 301-306.
34
References
DeltaCue Tutorial
This chapter presents a tutorial based on a small pair of DigitalGlobe
QuickBird images. These images have been co-registered (but not
cropped) and may be used to experiment with DeltaCue software.
The tutorial will step you through initial processing using the wizard
interface and change analysis using the DeltaCue change display
viewer. An example of iterating the processing parameters is also
included so that you can see a normal processing sequence from
start to finish. The tutorial quickly steps you through the three main
features of DeltaCue software:
•
Wizard Interface - described in detail in “DeltaCue Wizard
Interface” on page 53
•
Change Display Viewer - described in detail in “DeltaCue Change
Display Viewer” on page 69
•
Iteration Capability - described in detail in “DeltaCue Iterations”
on page 81
The tutorial data set consists of two small QuickBird images of the
area near Arâk, Iran, courtesy of DigitalGlobe. The goal of the
tutorial is to identify changes due to construction and changes in
agriculture. This process will illustrate typical usage of the DeltaCue
software, from initial processing to change analysis to iteration to
final product.
Data Set
The DeltaCue software installs two QuickBird images in the examples
directory of your IMAGINE software installation. Table 2 lists the
image names and sizes. The images have been co-registered and
remapped to a common pixel grid, but they have not been cropped.
Table 2: Tutorial Image Data Files
Image Name
Image Date
Image Size (pixels)
Pixel Size
DeltaCue.1.img
September 2002
1443 W x 1397 H x 4 bands
2.4 m
DeltaCue.2.img
July 2004
1443 W x 1397 H x 4 bands
2.4 m
Figure 10 shows the tutorial image pair in false color (4-2-1 band
combination). Note that the images have a large overlap area, but
they have not been cropped to their common extent. This cropping
operation is performed using the DeltaCue software. Also note that
the images do not have clouds within the common area. This fact is
used when specifying the image normalization process.
Data Set
35
Figure 10: Tutorial Images
Imagery courtesy DigitalGlobe
Initial Processing
Prior to using DeltaCue software, you must co-register the two
images from different dates. For this tutorial, that process has
already been performed on the example image data pair. Image coregistration is outside the scope of DeltaCue software, but it is a
critical preprocessing step. The pixel data in each must spatially
correspond to each other in order for DeltaCue’s change difference
process to perform properly. If you have data that has not been coregistered, you must manually co-register the data by either
remapping both to a common pixel grid (as was the case with the
example data) or remapping one image to the other. For this tutorial,
it is assumed that you have already co-registered the image pair and
you are ready to start using DeltaCue software to detect change
between the two dates.
Begin the change detection process by selecting DeltaCue icon in the
IMAGINE icon panel.
36
Initial Processing
The DeltaCue main menu dialog opens.
Normally, you always begin processing a new image pair using the
Wizard Mode in DeltaCue software. This interface helps you create a
project file and workspace directory for the image pair and
automatically runs DeltaCue processing workflow to create the initial
iteration. You then view that initial iteration using the DeltaCue
change display viewer. From within that viewer, you can select
different processing parameters for subsequent iterations on the
process.
Initial Processing
37
Click the Wizard Mode… button to open the following dialog.
Keep the default option to create a new project and specify a project
file name (for example, tutorial). The software automatically
appends the correct file extension (.dqw). If you do not select a
directory, your default data directory is used. The DeltaCue software
will create a workspace subdirectory with the same name as the
project file (minus the extension) in the same directory as the
project file.
Next select the Time 1 input image (DeltaCue.1.img) and the
Time 2 input image (DeltaCue.2.img).
NOTE: If you did not change your IMAGINE preferences for Default
Data Directory and Default Output Directory to the directory where
the tutorial data is located, click the file browse button to select the
file image files.
Select the sensor type to be QuickBird MS for both the Time 1 and
Time 2 images.
Keep the default settings option of Program Defaults.
When all the options have been specified, click the Next button is
enabled to move to the next wizard dialog.
38
Initial Processing
The second wizard page allows you to specify whether to crop the
image pair to a common area. This operation may or may not be part
of the co-registration process. In this example, the two images are
not cropped to a common area (see Figure 10), so you want to select
the option to crop the images.
When the DeltaCue process runs, it will produce cropped images in
the workspace directory. These images are called subset1.img and
subset2.img.
See “DeltaCue Workspaces” on page 101 for more information.
Click the Next button to move to the next wizard page. You may
cancel the wizard and the tutorial at any time. Note that no
processing is performed until the Finish button has been selected on
the final wizard page.
Initial Processing
39
The tutorial images do not contain clouds, but should be normalized.
The third wizard page allows you to specify whether the images
should be normalized and whether either image has clouds.
Generally, you should always normalize the images. DeltaCue
normalizes the Time 2 image to the Time 1 image. Clouds can skew
the normalization process so it is important to specify if either of the
images has clouds visible.
NOTE: Indicating that an image has clouds will not remove the
clouds from the change analysis, it will only adjust the normalization.
In this case, leave the cloud selection checkboxes unchecked.
Select Next to go to the next wizard page.
The fourth wizard page allows you to select the change algorithm to
be used. For this first iteration, we are looking for construction
activity and the magnitude difference algorithm generally does a
good job of identifying changes due to construction (new roads, new
buildings, and so on). Select the Magnitude Difference algorithm
and check the Interactive Thresholds box. Leave the Systematic
Change Threshold at the program default value of 30%.
The interactive thresholds setting allows you to interactively change
the set of four thresholds available when the process runs.
40
Initial Processing
Select Next to go to the change filter selection wizard dialog, shown
below.
The fifth wizard page is used for selecting change filters.
Select Spectral Segmentation but do not select other filters for
now. Generally, during your initial iteration, you want to see the
most change available. If those results are not acceptable, you can
apply a change filter in a subsequent iteration. Of course, if you know
that you are only interested in a particular size change, then you can
apply a spatial filter during this initial iteration.
Initial Processing
41
Select Next to move to the final wizard dialog.
The final wizard dialog allows you to remove selected material
categories of either the time 1 or time 2 input images from the
change detection result. For this initial run, do not remove any
materials from the output by leaving all check boxes unchecked.
Specify the output change image file name, as shown below. Enter a
name like “tutorial-1.img” (for the first tutorial iteration). For this
tutorial, we will not use a processing AOI so leave that option
unchecked.
Click the Finish button to begin processing. The DeltaCue change
detection workflow is run automatically at this point. A series of
progress dialogs similar to the one shown below is displayed as each
individual process within the workflow is executed.
You can force these progress dialogs dismiss automatically when the
process has completed by changing your IMAGINE preferences.
See Preference Settings on page 4.
42
Initial Processing
Since you elected to review and set the change detection thresholds
on the fourth wizard page, once the change has been computed the
software will display a viewer with the Time 1 image and the change
displayed as an overlay.
If the image does not have pyramid layers, an attention box is
displayed asking whether or not to compute pyramid layers.
A histogram of the change is also displayed (see Figure 11). This
dialog allows you to interactively change the upper and lower
thresholds and immediately see the amount of change remaining by
looking at the change overlay in the viewer. The histogram tool is
displayed on top of the viewer that contains the thresholded change
image so you should move the histogram window to the side.
The histogram plot consists of two tabbed displays, one for the
Lower Thresholds and one for the Upper Thresholds. Lower
thresholds control the cyan overlay and upper thresholds control the
yellow overlay. Triangle controls at the bottom of the plot allow you
to adjust the threshold values. There are two upper and two lower
thresholds. Only change values between the two sets of thresholds
are retained. You can control the horizontal scale of the histogram
plot using the Range pull down control.
The default outer threshold and range values prevent both upper
or lower triangle controls from being displayed initially.
The change values in the histogram are relative difference values
(fractions) that have been scaled by a factor of 1000. The total
possible range for the change values is ±32767 which corresponds
to the range of signed 16-bit image data. Thus, the initial 30%
threshold corresponds to a change value of 300 (0.3 x 1000). The
outermost threshold values are initially set to the extent of the
change histogram.
As you adjust the lower (cyan) thresholds using the triangle controls
at the bottom of the histogram, the amount of negative (cyan)
change will vary in the viewer. If you select the tab for Upper
Thresholds, you can adjust the upper thresholds and the amount of
positive (yellow) change in the viewer.
Initial Processing
43
Figure 11: Time 1 Image and Histogram
Try varying the thresholds and note the impact on the amount of
change left. You can also adjust the threshold values using the text
boxes.
For more information on setting thresholds refer to "Setting
Interactive Change Thresholds" on page 65.
For the purposes of this tutorial, reset the thresholds using the
Reset button beneath the histogram display and then select Finish
to accept those values and complete the process. This will reset the
thresholds to their initial 30% values and continue with the DeltaCue
workflow.
When you click Finish, the Viewer and the Histogram windows are
closed.
44
Initial Processing
Change Analysis
Once the DeltaCue workflow completes, the software automatically
launches the change display viewer to help you analyze the detected
change. An example is shown on the next page. Use the “zoom in by
2” control
and the roaming control
to zoom in on the area
shown. Note that both sides of the viewer act in unison. If the two
sides should get out of sync with each other in terms of zoom level,
you can use the
control to resync the views. Detailed usage of
the change display viewer is covered in DeltaCue Change Detection
on page 23.
The change pixels are color-coded according to their before, after, or
transition classes. To see the color coding better, select the
button to remove the background image on the right side. The set of
before classes are shown here.
To see an example of new construction, select the change magnifier
and select a yellow pixel in the right hand display. Blue areas are
areas where the overall magnitude of the pixel value is brighter in
Time 2 than it was in Time 1 and the before class material was soil.
Magnitude changes are frequently associated with significant
changes in the landcover material, such as a new road or new
building. The example above shows new road construction leading to
a tunnel. By selecting other points within the image you can see
other areas where change has occurred.
Change Analysis
45
A second possible iteration at this point would be to call up the
DeltaCue iteration menu and select a spatial filter to eliminate small
clutter or the misregistration filter to eliminate unwanted change due
to local misregistration between the two images.
For the purposes of this tutorial, however, at this point we will turn
our attention to agriculture. The magnitude difference algorithm
does not necessarily do a good job of detecting changes in
agriculture because the overall spectral magnitude may be the same
between the two dates. An example of this effect is shown below.
In this example, the field marked is displayed in Time 1 but not in
Time 2 and the change has not been indicated. The magnitude
difference is not especially sensitive to vegetation changes. A better
algorithm to use would be the Tasseled Cap greenness difference
which is better tuned to vegetation changes. As a second iteration,
we will reprocess the image using the Tasseled Cap greenness
difference to indicate changes in vegetation.
Iteration
46
Select the DeltaCue iteration control
as shown below.
. The iteration dialog opens,
Iteration
This dialog consists of three tabs: one for change algorithms, one for
change filters, and one for material filters. The tab for change
algorithms is shown. Select TC Green Difference as the change
algorithm.
Enter an output file name such as tutorial-2.img (for the second
iteration image). Select OK to begin processing for the second
iteration. The DeltaCue software will run only those portions of the
process that it needs to produce an output. Since image cropping
and normalization have already been run and the results are stored
in the workspace file, those processes will not be repeated.
Once again the threshold program will display a viewer with the Time
1 image and the change overlay (which is different from the first
iteration). It will also display the change histogram and interactive
controls. Experiment with changing thresholds to see their effect on
the amount of change left. For the purposes of this tutorial, select
the Reset button below the change histogram plot and select Finish
to complete processing.
Once processing has completed, the change display viewer is
updated with the new iteration results. The change layer on the right
hand side will now contain the change detection results for the new
iteration.
NOTE: The previously undetected field change is now clearly
detected.
Iteration
47
The Tasseled Cap difference is a sensitive indicator of changes in
vegetation. The change in the agricultural field that was missed by
the magnitude difference is clearly detected by the Tasseled Cap
greenness difference.
To see what else was detected, use the Zoom All button
to zoom
the display to the full extents of the image, as shown below.
48
Iteration
Zoom to
this area.
Zoom into the area shown above. This area contains numerous
agricultural fields in various states of bare soil and vegetative growth
at Time 1 and Time 2. As shown below, the blue colored fields
represent a change from dark disturbed soil in Time 1 to something
else in Time 2. We will next use the spectral segmentation tools to
filter out all but very specific types of change.
Iteration
49
Assuming we are only interested in changes from dark disturbed soil
to bright, lush vegetation, begin by eliminating the unwanted before
classes.
Right-click in the left view and select Arrange Layers from the
Quick View context menu. In the Arrange Layers dialog, right-click
on tutorial-2.img and select Attribute Editor from the
PseudoColor Options menu. Scroll down in the Raster Attribute Editor
to row 25 and click in the Row column to select it. Hold the shift key
and select the remaining 3-x classes (rows 25-36). Right-click in the
Row column and select Invert Selection from the Row Selection
menu. Left-click the Opacity column head to select all in this column
then right-click and select Formula from the Column Options menu.
Click in the Formula window of the Formula dialog, enter 0, and click
Apply.
Select the Off control
to turn off a selected class using the
mouse. Select the On control
to turn a class back on. Classes
selected for elimination will turn white to indicate that they will be
removed when you select Apply. Selecting the Reset button
restores all classes. Begin by turning off all before classes that are
not colored blue (see following page).
50
Iteration
Now select the “After” state and turn off unwanted after classes. You
may fine-tune the result by turning off unwanted transition classes.
The resulting spectrally filtered change detection result is shown
below. All instances of these change pixels throughout the image
represent changes from dark disturbed soil to lush active vegetation.
FPO
Iteration
51
A possible new iteration on this result would be to apply a spatial
filter to eliminate fields based on their area or spatial extent. The size
parameter selected would depend on your particular application. The
misregistration filter might also be applied to eliminate superficial
apparent change along the edges of fields.
52
Iteration
DeltaCue Wizard Interface
Introduction
The DeltaCue wizard interface is the primary method of using
DeltaCue software to produce an initial set of change detection
results. Once you have produced initial results, you can view them
with the DeltaCue change display viewer and run new processing
iterations from within the change display viewer. The wizard
interface steps you through a sequence of dialogs that capture all of
the necessary inputs to run the DeltaCue process end-to-end. The
software creates a project settings file and a workspace directory to
hold intermediate files. When the wizard process completes, the
DeltaCue change display viewer program is automatically launched
so that you can view and analyze the change detection results.
See “DeltaCue Change Display Viewer” on page 69 for more
information about the special DeltaCue change display viewer.
Once you have created a project file using the wizard interface, you
can then run different process iterations to fine-tune the results.
See “DeltaCue Iterations” on page 81 for more information
about iterating with the DeltaCue process.
Initial Wizard
Usage
The DeltaCue wizard interface is a typical wizard-style user interface
that is designed to step you through a sequence of dialogs to capture
information needed to run the DeltaCue process. Next and Back
buttons control stepping through the various wizard interface
dialogs. The program will not let you move to the next dialog until all
the required inputs for the current dialog have been entered. The
Finish button is used to actually launch the processing and the
program will not let you select the Finish button until all required
inputs have been entered. The Cancel button can be used at any
time to cancel the process. Note that if you cancel the process all
user inputs entered up to that point are lost.
Normally when you first use DeltaCue software to process an image
pair for change, you will place the images in a working directory and
co-register them using IMAGINE AutoSync or some other tool. The
two input images must be co-registered before you begin the
DeltaCue wizard process. The image pair does not have to be subset
to a common area since DeltaCue will perform this operation if
specified.
When you are ready to begin the DeltaCue process, select the
Wizard Mode… button from the DeltaCue main menu.
Initial Wizard Usage
53
Project Selection Dialog
The DeltaCue wizard interface project selection dialog is shown here.
This dialog is used to start the creation of a new workspace file (.dqw
file) or select an existing project to which you want to add a new
iteration. The default setting is to create a new project file. DeltaCue
project files store the settings used for a processing run and allow
you to view results and iterate on them. Project files are XML files
that may be viewed (and carefully edited if needed) so that you have
a history of the processing applied to achieve a certain result. This
section describes the initial use of the wizard to create a new project.
Refer to the next section if you want to use an existing project file
with the wizard interface.
Leave the radio button selected on Create a New Project to create
an initial project file for your image pair. The required inputs on the
first dialog of the wizard are:
54
•
Name of output project file
•
Name of Time 1 image file
Initial Wizard Usage
•
Time 1 image sensor type
•
Name of Time 2 image file
•
Time 2 image sensor type
You must specify an output project file name. Use the file browse
icon
to control where the project file will be located. The
DeltaCue process will create a workspace directory to hold
intermediate process files in the same path as the selected project
file and will have the same root name as the project file.
You must have write permission to the project directory and the
disk must have sufficient free space to hold the intermediate
files produced, typically at least 1 GB.
Select two existing image files using the file browse icon. These
images do not have to be located in the same directory as the project
directory, but you must have write permission to the directory where
the images are stored. The DeltaCue process will create image
attributes auxiliary files (.atr files) in the directory where the images
are located.
For each image, select the corresponding sensor type. If the sensor
type for your imagery is not listed, select Other. The sensor type
setting is used to determine whether Tasseled Cap coefficients are
available for the image. Selecting Other as the sensor type will
disable those features that rely upon the existence of a Tasseled Cap
transform for the image. If your sensor is panchromatic (only one
band of data per image), then all of the change algorithms are turned
off with the exception of the single band difference algorithm as the
other algorithms are designed for multispectral data.
DeltaCue software provides the ability to save program settings and
use them for future processing. The optional Settings pull-down
menu allows you to select a previously stored set of processing
settings. These settings are applied in subsequent wizard dialogs as
defaults. You may change them on those dialogs.
See “DeltaCue Iterations” on page 81 for more information on
how to create a settings file.
Image Cropping Dialog
Initial Wizard Usage
Once you have specified the required information on the first wizard
dialog and select Next, the second wizard dialog opens, as shown
here.
55
DeltaCue change detection processing requires that the coregistered image pair overlap exactly. All pixels in each image
must be cropped to a common area and pixel size for both
images. Images that mostly overlap must be cropped even if
there is only one row or column of difference between the two
images.
The process provides the ability to crop the two images down to a
common area and pixel size. If the image pair has already been
cropped as a result of the co-registration process, then you should
select No, don’t crop the images which is the default. Otherwise
select Yes, crop the images and the process will automatically crop
the image pair to a common area subset and pixel size as a first step
in the processing chain.
Normalization Dialog
56
The third initial wizard dialog, shown on the next page, controls the
image normalization process within DeltaCue software. This process
creates a normalized Time 2 image with statistics that match that of
the first Time 1 image. It is important during this process that clouds
not skew the results. Therefore you should indicate whether each
image contains clouds or not. The default setting is that the images
do not contain clouds and all pixels are used during image
normalization. If an image contains clouds, an unsupervised
classification process is used to identify the brightest classes and
these are excluded from use in normalization. This unsupervised
classification process adds time to the overall processing time, but is
required if either image has substantial cloud cover.
Initial Wizard Usage
Specifying that an image has clouds does not mask the clouds
from the change detection process. It merely limits their
influence during the normalization calculation.
If the images have previously been radiometrically corrected or
normalized or you are confident that the two images are statistically
similar (due to similar atmospheric and illumination conditions), you
may skip the image normalization process in DeltaCue.
Change Detection Dialog
Initial Wizard Usage
The Change Detection wizard dialog, shown here, is the main change
detection selection dialog. This dialog controls which change
detection method is used to produce the final change detection
output.
57
You should begin by selecting the desired change algorithm (or
keeping the selection specified in the settings file you selected
initially).
Two forms of Tasseled Cap transforms are available for change
detection. Tasseled Cap Greenness Difference measures changes
primarily in vegetation while Tasseled Cap Soil Difference measures
changes in soil and other non-vegetative materials. With either of
these algorithms, you must specify a sensor on the first wizard dialog
since these transforms are sensor specific. If the sensor for your
imagery is not listed, you cannot use a Tasseled Cap change
algorithm with your data. In that case, select eitherMagnitude
Difference which detects change based on spectral magnitude
differences or a single-band difference.
The Tasseled Cap change algorithms are disabled if the sensor
type was specified as OTHER. Also, if processing panchromatic
imagery, only the single band difference algorithm is available.
If you select one of the primary color difference algorithms (red,
green, blue diff), you must also specify the color threshold as a
percentage. This percentage applies to the spectral color cosine
value before the difference operation is applied.
See "Change Algorithms" on page 25 for more information.
58
Initial Wizard Usage
Thus, a threshold of 65% means that only those cosine values that
exceed 0.65 are included in the difference. Otherwise, the value is
set to zero. This insures that the difference only shows changes
related to the primary color selected.
To loosen the criteria and let other colors possibly affect the result,
lower the threshold. Increasing the threshold will restrict the
resulting changes further so that only very pure primary colors will
affect the change difference.
Once you have specified a change detection algorithm you must
specify the threshold percentage. The DeltaCue process uses the
histogram of the change detection output and thresholds that
histogram using four thresholds as an initial way of eliminating
unwanted change. An example of a change difference histogram is
shown below.
-30%
Lower
Thresholds
+30%
Upper
Thresholds
For some applications a simple plus and minus percentage threshold,
such as the one shown above may be sufficient to eliminate most of
the unwanted change in the central peak of the histogram. All
change greater than the positive percentage or less than the
negative percentage is carried forward to other filters. In that case,
specify a percentage amount and leave the box entitled Interactive
Thresholds unchecked.
Initial Wizard Usage
59
If you need finer control over the initial threshold, check the box
entitled Interactive Thresholds. This will cause the DeltaCue
threshold program to present an interactive display of the histogram
showing the thresholds and an image display with a change area
overlay. As you adjust the thresholds, the change overlay changes
in response to show you the effect of each threshold. Four
thresholds, two upper and two lower, are used. Only those change
values that lie between the two upper and two lower thresholds are
retained for further processing.
For more information on the interactive threshold program, refer
to "Setting Interactive Change Thresholds" on page 65.
Change Filters Dialog
The fifth wizard dialog allows you to select change filters. Change
filters are also used to further eliminate unwanted change.
Three types of changes filters are controlled by this wizard dialog:
60
•
spectral segmentation
•
misregistration
•
spatial filtering
Initial Wizard Usage
Spectral Segmentation
Spectral segmentation is a process that is applied to the Time 1 and
Time 2 images to classify the change pixels into spectrally similar
classes. In the change display viewer, you can then interactively
filter out change pixels based on their before or after spectral class.
If you want a simpler output that merely indicates positive and
negative changes, uncheck the Spectral Segmentation option and
the process will skip classifying the Time 1 and Time 2 change pixels.
This saves processing time, but the resulting change image only
indicates whether the brightness change was positive or negative,
and not the spectral class before and after.
Misregistration
The misregistration filter attempts to filter out unwanted change due
to local misregistration of the image pair. Such pixel misregistrations
can cause apparent change differences simply because the correct
pair of pixels was not differenced.
The misregistration filter checks the local neighborhood surrounding
each pixel identified as having changed. The size of the search
window is specified in the wizard dialog, as shown on the next page.
The filter will search within the search window to see if another pixel
in that window would satisfy the criteria for no significant change. If
a match is found, then that pixel is a candidate for being due to
misregistration.
Since local misregistrations typically occur in a shift-like pattern, the
process tries to validate the misregistration candidate by examining
nearby neighbors to see if they follow the same pattern of no change.
The number of nearby neighbors considered is controlled by the
search window size. This parameter may be reduced to reduce run
time if needed.
Spatial Filtering
The spatial filter identifies contiguous blobs of detected change by
delineating the contour of the blob. A contiguous blob is a set of
pixels that are connected by at least one neighbor in any of eight
directions. Two change areas are connected if they share at least one
neighbor in common.
Initial Wizard Usage
61
Once the spatial filter process has detected the contour of a
contiguous change area, it computes several geometric properties
based on the contour.
The geometric properties considered are:
•
area
•
major axis length
•
minor axis length
•
compactness
•
elongation
The definition of each geometric property is provided in Table 3.
These geometric properties can be used to filter out unwanted
change using a range of values.
Table 3: Geometric Properties Used for Spatial Filtering
Property
Definition
Area
Total area of the shape
Major Axis Length, P1
Length of line segment intersecting shape
along major axis of rotation (symmetry)
Minor Axis Length, P2
Length of line segment intersecting shape
along minor axis of rotation (symmetry)
Compactness
A ,
C = ----------P1 P2
Elongation
P
E = -----1 ,
P2
A = Area
P1 = Major Axis Length
P2 = Minor Axis Length
P1 = Major Axis Length
P2 = Minor Axis Length
Area, Major Axis, and Minor Axis length have units associated
with them. Select the Units from a dropdown menu.
For example, if the units are meters, the area is in square meters and
the lengths are in meters.
Compactness and Elongation are dimensionless quantities.
To filter a geometric property, check the checkbox associated with
that property and specify the minimum and maximum values to
keep.
NOTE: Minimum and maximum range is inclusive, therefore both the
minimum value and maximum values are included and their
corresponding shapes retained.
62
Initial Wizard Usage
Thus, if you specify a minimum and maximum area of 10 pixels, only
shapes with exactly 10 pixels are retained. If the maximum area is
set to 11 pixels, then shapes with both 10 and 11 pixels are retained.
However, if the minimum area is set to 10.1 pixels, then only shapes
with 11 pixels are retained.
Material Filter Dialog
Once you have specified the change algorithm, threshold parameter,
and change filters, you may move on to the final dialog in the wizard
interface, shown below.
This final wizard dialog allows Material Filtering to be specified.
Material Filtering is only available for sensors that have Tasseled Cap
coefficients derived for them. If you selected one of the sensor types
to be OTHER on the first wizard dialog, then the Material Filtering
section is grayed out since Tasseled Cap coefficients are not
available for that sensor pair.
Material Filtering allows you to exclude selected categories of
materials from the change detection result.
For example, by clicking the checkbox for Vegetation in Time 1,
you are electing to exclude any change that includes vegetation in
Time 1 category. Changes that involve vegetation in Time 2 are not
excluded unless you check the Time 2 Vegetation checkbox.
Material Filtering is intentionally conservative. The process
generally does not exclude change pixels that include mixtures
of material categories since those may be of interest. The
material filter only applies to those pixels that clearly fit the
category. Other change pixels will have to be filtered out by
other means.
Initial Wizard Usage
63
Once you select a Material Filter category by checking the checkbox
next to the category in either Time 1 or Time 2, the parameters
associated with that material filter become enabled. The default
values generally apply since they are conservative.
Refer to “DeltaCue Material Filtering” on page 105 for a
description of the Material Filtering process.
Session Output
In the Material Filtering dialog, the Session Output section is the
final output file selection and a mechanism for selecting an AOI to
use during processing.
Specify the name of the output image file that will contain the filtered
change detection results. This image is a color coded raster image in
which the pixels indicate the type of change. The DeltaCue change
display viewer is used to view this change image. The name is stored
as part of the DeltaCue workspace file for easy retrieval in the
change display viewer.
See “DeltaCue Workspaces” on page 101.
If you would like to process only a portion of the input image pair,
you may create an IMAGINE Area of Interest (AOI) and save that AOI
to a file.
In this dialog, check the checkbox labeled Use AOI File. This
enables the AOI file selection box. You may then select the AOI file
that you previously saved. The AOI must be in a saved file and not
just a temporary AOI created in a viewer. If you have a viewer with
an AOI that you would like to use, first save that AOI to a file and
then select it.
Once you have specified all required inputs, including the output
image file name, the Record and Finish buttons at the bottom of
the wizard dialog become active.
The Record button allows you to record the inputs in the previous
wizard dialogs to the DeltaCue workspace file. These inputs would
then be used to create an output iteration when you eventually run
the process.
64
Initial Wizard Usage
Recording an iteration allows you to click the wizard Back button to
go back to previous dialogs and specify a different set ofprocessing
parameters. That set of parameters would then constitute a new
iteration. When you finally select the Finish button, all recorded
iterations are run during a single set of processing sequences. If you
select the Finish button without selecting Record, then the process
is run with only one iteration based on the parameters that you
selected.
See “DeltaCue Iterations” on page 81 for more information on
iterations.
When you click the Finish button, a project file (.dqw file) is created
and a separate application, called dQrunprocess, takes over. This
program reads the project file just created and controls execution of
the DeltaCue change detection process based on the settings stored
in the project file. A number of separate processes are spawned to
create intermediate outputs. Each separate process will create its
own progress dialog. When all of the processes in the processing
chain have completed you will see the dQrunprocess progress dialog
indicate completion.
If you modified your IMAGINE preferences to close all progress
dialogs, then all progress dialogs will close automatically. Otherwise,
you must close them by selecting the OK button on each progress
dialog that was started. It is also a good idea to check the session
log for any unusual occurrences or error messages.
As a final step, the dQrunprocess program automatically spawns the
DeltaCue change display viewer program to allow you to see the new
change detection results and possibly iterate on them.
Refer to “DeltaCue Change Display Viewer” on page 69 for
information on how to use the change display viewer.
Setting Interactive
Change Thresholds
Initial Wizard Usage
The DeltaCue process is designed to be largely automatic. There is
one optional step. If you selected the checkbox for Interactive
Thresholds on the Change detection wizard dialog, the interactive
version of the DeltaCue change threshold program is run.
65
This process applies the initial threshold percentage specified and
allows you to interactively set four thresholds, two upper and two
lower.
The process first displays the Time 1 image in a viewer with a colorcoded thresholded change image overlay. In the viewer, the change
area overlay is the top layer with the Time 1 image beneath it.
Positive change is colored red and negative change is colored blue.
A separate window is displayed with the change detection histogram
plotted. An example is shown below.
The histogram window contains two tabbed plotting windows, one for
the lower thresholds and one for the upper thresholds.
When the Lower Thresholds tab is selected, you may change the
lower thresholds by moving the triangular symbols at the bottom of
the histogram plot.
When the Upper Thresholds tab is selected, you control the upper
thresholds by moving the triangle symbols at the top.
Another method to change the threshold values is to change the
number box for each threshold.
When you change a threshold value, the change overlay in the
viewer automatically updates to show the area included within the
upper and lower thresholds. Increasing the extent between the
threshold pairs increases the amount of image area covered by
change. This change area is passed on to subsequent filters or to the
final change output.
NOTE: Right-mouse click on the threshold number boxes to set the
threshold increment value when you press the up and down arrows.
66
Initial Wizard Usage
Note that the initial histogram Range displayed is +/- 1000. The
histogram is scaled by a factor of 10 so this nominally represents +/100% change. However since some change detection algorithms can
exhibit higher amounts of relative change, a Range dropdown menu
is provided to change the scale of the histogram. If a threshold
control value falls outside the displayed range, it will not be available
for adjustment. Adjust the inner thresholds at finer range and use
larger ranges for the outer thresholds as needed.
Wizard Usage
with Existing
Project Files
Normally, you would iterate through different processing settings
using the DeltaCue iteration capability.
See “DeltaCue Iterations” on page 81.
However, you can rerun the DeltaCue wizard and select an existing
project file to create a new iteration or set of iterations, as shown in
the dialog below.
In that case, the settings from the last iteration of the project are
used as the default. Since this is an existing project, the dialogs
asking about common area subsetting and image normalization do
not apply and these dialogs will not appear. Those steps have
already been performed on the image pair in the project file.
Likewise, DeltaCue settings do not appear at the bottom of dialogs
since these have already been applied and possibly modified. The
settings in use are those of the last iteration for that project file.
1. Enter the name of the existing project or click the file browse icon to
locate the project.
Wizard Usage with Existing Project Files
67
2. In the Change detection dialog, modify the settings shown which
reflect the settings used during the last iteration.
3. In the Change Filters dialog, modify the settings shown which reflect
the settings used during the last iteration.
4. In the Material Filtering dialog, specify a new change detection
output image name.
This is normally a new file but you may overwrite an existing output
file. The interface will prompt you before it overwrites an existing
file.
5. Click Finish.
The project file is updated with a new iteration and the dQrunprocess
program is spawned to conduct the new processing steps. If existing
intermediate project files exist, these will not be overwritten. The
process will assume that you want to reuse these files and only
create unique new files. This saves processing time when you only
want to adjust a change filter or its parameters.
Once the new change detection output file is create, the DeltaCue
change display viewer program is launched so that you can view the
new results. This process is similar to DeltaCue iterations, but the
iterations mechanism is more flexible and allows you to save
DeltaCue settings files for future use by yourself and others.
See “DeltaCue Iterations” on page 81.
68
Wizard Usage with Existing Project Files
DeltaCue Change Display Viewer
The DeltaCue change display viewer is a specialized IMAGINE viewer
application specifically designed for change detection analysis. The
viewer displays the original image pair as well as the change
detection output. Tools appropriate for change detection
interpretation and analysis are made readily available. The viewer
accepts DeltaCue project files as input and provides an interface to
the DeltaCue iteration capability. Using this capability you can
perform different iterations on the processing to determine which
settings are best suited for your particular application. You can save
those settings in a DeltaCue settings file (.dqs file) for later use with
other image pairs.
Viewer Features
By default, the DeltaCue change display viewer displays two views
side by side. The left side view displays the original image pair as a
layered stack with the Time 1 image on the top (visible) and the Time
2 image on the bottom. The right-side view displays the Time 2
image on the bottom and the color-coded change detection image on
the top. An example is shown on the next page. Note that the change
pixels are color-coded according to spectral classes while the nonchange pixels have an opacity of 0, allowing the underlying time 2
image to be seen. You can view the change classes from time 1
(before classes) or time 2 (after classes) or the transitions from time
1 to time 2. You can interactively select these classes for removal
from the change overlay.
Change
Overlay
Viewer Features
69
You can zoom and pan around the image area using familiar
IMAGINE viewer controls. These are described in more detail in
subsequent sections. The two windows within the main viewer are
linked with each other so that they change simultaneously, with one
notable exception – the zoom tool buttons (
and
). When
using the zoom buttons, only one window zooms in or out. The other
window shows the geographic area covered in the zoomed in
window. A control button is provided to re-synch the views following
a zoom operation using the zoomin or zoomout buttons. This is
described in more detail below.
Layer swipe and flicker are two of most useful tools in IMAGINE for
visualizing change between two images. The DeltaCue change
display viewer provides convenient access to both tools applied to
the two original images loaded in the left side of the viewer. An
example of layer swipe used to visualize change is shown below. In
this example, the detected change is shown in yellow and red on the
right and layer swipe is used to peal back the top Time 1 image layer
on the left to reveal the state at Time 2. The change between the two
times is evident in the process. The flicker control, available as a
button on the display controls, may also be used to switch back and
forth between Time 1 and 2 images on the left to see the overall
change between dates.
Layer
Swipe
70
Viewer Features
A special change magnifier has been added to the DeltaCue change
display viewer to help locate and analyze change. An example is
shown below.
Change
Magnifiers
When the magnifier button is turned on, two additional smaller
display windows appear within the top portion of the left and right
viewers. As you use the mouse to select locations within the lower
main image, those locations appear in magnified form in the
additional magnifier windows. The Time 1 image is displayed on the
left magnifier window and the Time 2 image is displayed on the right.
The difference between the two dates at that location is readily
evident. You can control the appearance of the magnifier windows
through a change magnifier properties dialog, shown here.
The classic IMAGINE viewer measurement and cursor tools are also
available and function in both the left and right side of the change
display viewer.
Viewer Features
71
Viewer Menus
The DeltaCue change display viewer menus provide functionality for
manipulating files, calling up utility functions, changing the viewer
properties, invoking special tools, and online help, as shown here.
The File menu allows you to open and save files, print views, clear
or close the viewer, and create new layers. You may only create new
AOI, annotation, or shapefile layers using the File | New menu.
These layers are overlaid on the existing raster layers in each view.
The File | Open menu allows you to open a change detection
project, and AOI, annotation, and shapefiles layers. The selected
files are opened and overlaid in both the left and right side views and
they may be manipulated separately.
The File | Save menu allows you to save left and right AOI,
annotation, and shapefile layers to new files. You may create new
layers, edit them with their associated tool sets, and then save them
to files for later use.
The File | Clear button clears all layers in both the left and right
views. This effectively resets the viewer. You may clear separate
AOI, annotation, and shapefile layers using the Arrange Layers
control described below.
The File | Close button terminates the viewer program.
The Utility menu provides access to two utility functions: the Inquire
Cursor and the Measure tool. These utilities have separate standard
IMAGINE interfaces and they apply to both the left and right views.
72
Viewer Menus
The Tools menu allows you to call up tool sets for the three types of
overlays available with the viewer: AOIs, annotation layers, and
shapefile layers.
These tools sets are standard IMAGINE tool sets for each layer type
and are shown here. The vector tools apply to shapefile layers, which
are the only vector layers used in the DeltaCue change display
viewer. Refer to IMAGINE documentation for the usage of each tool.
Annotation
Tools
Vector
Tools
AOI
Tools
Viewer Menus
73
Viewer Controls
The DeltaCue change display viewer controls are a set of icon
buttons related to functions that are frequently used in change
detection analysis. Many of these controls are standard IMAGINE
viewer controls while some are new to this viewer.
The first row of viewer controls consists of the following:
File Open – this control brings up the dialog for opening a
change detection set, as shown on the right. You may specify either
a DeltaCue project file (.dqw file) or specify the individual inputs.
Unlike the regular IMAGINE viewer, this control is only used to open
raster layers. Use the File | Open menu to open overlay layers such
as AOIs, annotation layers, and shapefile layers.
Clear Viewer – this control clears the viewer and removes all
raster and overlay layers from both sides of the display.
Reset Zoom – this control resets the zoom level in each view
to a default level centered about the center of the images. Both sides
of the viewer display are reset simultaneously to the same default
zoom level and extent.
Zoom In By 2 – this control zooms both sides of the display in
by a factor of two about the current center of each image. Both sides
zoom in simultaneously and retain their relative views.
Zoom Out By 2 – this control zooms both sides of the display
out by a factor of two about the current center of each image. Both
side zoom in simultaneously and retain their relative views.
Zoom All – this control zooms both sides of the display out to
the full extent of each image.
74
Viewer Controls
Synchronize Zoom – this control re-synchronizes the zoom
levels and extents of each side of the display. Occasionally one side
may become more zoomed than another side and a box is displayed
to show the extent of the side that has the higher zoom level. This
control can be used to re-synchronize the two sides to the same
zoom level. The side with the higher zoom level is used as the basis
for synchronizing the zoom level.
Cursor Tool – this control invokes the IMAGINE cursor tool.
This tool may be applied to either side by simply moving the crosshair in that side. The information in the cursor window applies to the
selected side. An example of the cursor tool is shown below. Note
that cross-hairs are displayed in both views and are linked.
Selection Tool – this is the default selection tool. For raster
layers it does nothing, but for overlay layers it can be used to select
objects. This tool is used to turn off the other tools such as the zoom
and pan tools.
Zoom In Tool – this control is used to zoom in the display. You
may zoom in by a factor of two by clicking in a window or you can
select the zoom extent by dragging the zoom tool cursor within a
view window. Occasionally the other side of the display will not zoom
to the extent selected. Use the
tool to re-synch the displays.
Zoom Out Tool – this control is used to zoom out the display.
You may zoom out by a factor of two by clicking in a window or you
can select the zoom extent by dragging the zoom tool cursor within
a view window. Occasionally the other side of the display will not
zoom to the extent selected. Use the
displays.
Viewer Controls
tool to re-synch the
75
Pan Tool – this control is used to pan one side of the display.
The other side automatically follows the panning action to keep the
two sides of the display in synch.
Contrast/Brightness Adjustment Tool – this control is used
to adjust the brightness and contrast of the time 1 and time 2
images. The control on the left brings up the simple
contrast/brightness adjustment tool in IMAGINE. This tool may be
used to adjust the contrast and brightness of the top image in the
left-hand display. Use the Flicker tool
layers in the left hand display.
to switch the order of the
Contrast/Brightness Update Tool – this control applies your
contrast and
brightness changes, made in the left hand display
using the Contrast and Brightness control, to all views, including the
change magnifier views. Use this control once you have adjusted the
contrast and brightness for the layers in the left-hand display.
DeltaCue Iterations – this control invokes the DeltaCue
iteration dialog.
See “DeltaCue Iterations” on page 81.
DeltaCue Iteration Selection – this control allows
you to change the currently displayed iteration without reloading the
entire project. When you change the iteration selection, the change
detection layer displayed on the right side changes to correspond to
the selected iteration.
Shapefile Output - this control converts the current change area
to a shapefile (.shp) and writes it to disk. You are first prompted for
an output file name. The shapefile is automatically displayed in both
views.
76
Viewer Controls
The second row of viewer controls is as follows:
Zoom Tool – this control allows you to
adjust the zoom level in both sides of the display in small
increments. The thumbwheel control can be used to adjust the zoom
level up and down interactively. The negative zoom control zooms
the display out by a small increment and the positive zoom control
zooms the display in by a small increment. You can also reset the
zoom level with the reset control on the far right side.
Layer Swipe Tool – this control brings up the layer swipe
control panel shown below. The swipe tool allows you to pull back the
top Time 2 layer on the left hand side to reveal the Time 1 image
beneath. By swiping back and forth you can see differences between
the two images. The boundary of the top layer is controlled by the
swipe position slider bar in the swipe control panel. As you move this
slider back and forth, the amount of Time 2 image visible on top
changes. You can also change the direction of the swipe action and
toggle an automatic swipe mode in which the left display continually
swipes back and forth at a specified speed. When the swipe tool is
cancelled, the display returns to normal.
Flicker Tool – this control toggles the order of the Time 1 and
Time 2 layers in the left hand display to simulate a flicker control.
When the Flicker Tool is in the depressed state, the Time 2 image is
visible. Otherwise, the Time 1 image is visible. By repeatedly
toggling this layer on and off you can see differences in the Time 1
and Time 2 images and correlate them with the change detection
results shown in the right side display.
Viewer Controls
77
Measure Tool – this control invokes a pair of standard
IMAGINE measurement tool sets as shown on the next page. The
tool set for viewer #1 applies to the left side display and the tool set
for viewer #2 applies to the right side display.
The measurement tool set allows you to measure coordinate
positions, line lengths, areas, and perimeters in a number of
different ways. You can save your measurements to a text file which
can also be viewed by the measurement tool. Refer to the IMAGINE
user documentation for more information on the features of this tool.
North Symbol – this control toggles a north arrow on and off
in both the left and right side displays. You can reposition the north
arrow by selecting it with the mouse and dragging it to a new
position. The north arrow is automatically repositioned to the same
location in both the left and right views.
Scale Symbol – this control toggles a scale symbol on and off
in both the left and right side displays. The scale symbol provides a
horizontal and vertical distance bar that can be moved around and
compared with features in the image. You can reposition the scale
symbol by selecting it with the mouse and dragging it to a new
position. The symbol is automatically repositioned to the same
location in both the left and right views.
78
Viewer Controls
Change Magnifier – this control toggles a special pair of
change detection magnifier windows on and off. The magnifier
window in the left side contains a magnified view of image data from
Time 1 and the window in the right side contains the corresponding
view of image data from Time 2. By comparing the two magnified
images you can readily see changes that have occurred at that
location. You cannot reposition the change magnifier windows, but
you can change their size and magnification properties with the
magnifier properties tool described below.
Set Center - this control is used to set the center of the
magnified view. Click this control then click in the normal view at the
location on which you would like the magnified view to be centered.
Magnifier Properties – this control brings up the change
magnifier properties dialog shown below. This dialog can be used to
change the height of the magnifier windows and the magnification
level applied. Magnification levels can range from a factor of 1 to 8.
Window height is expressed in terms of screen pixels. You can also
change the color of the cross-hair symbol for better visibility.
Change Background – this control toggles the visibility of the
background layer on the right side view so that you can more readily
identify change areas. The icon changes to remind you that it is in
effect.
Viewer Controls
79
Spectral Filtering – these controls allow
you to interactively filter detected change based on the spectral
characteristics of the before and after landcover materials or the
transition between materials. Select the overlay to be displayed
using the radio buttons for Before, After, and Transition. The
display will change to show color-coded spectral classes associated
with Before, After, or Transition classes. Select the Off control
to interactively turn off a selected class using the mouse. The
selected class will change color to white to indicate that it has been
selected. Select the On control
to interactively turn a selected
class back on. The type of classes displayed and selected will depend
on the radio button selected. The white classes selected for
elimination are removed when you select Apply.
Once you are satisfied with your selections, select Apply to eliminate
the selected classes from the display. Note that those classes are not
eliminated from the image, simply from the display. Select Reset to
restore all classes.
Spectral Bands – these controls
allow you to change the band combinations that are associated with
the red, green, and blue channels of the computer display. The Time
1 and Time 2 images in each side of the display automatically change
when you change the band combination in effect.
Right-click in a viewer to display a Quick View context menu of
additional functions.
80
Viewer Controls
DeltaCue Iterations
DeltaCue software provides the capability to easily try different
processing settings to find the ones that provide the best results for
your imagery and your application. Once you have created an initial
project file using the DeltaCue wizard (see “DeltaCue Wizard
Interface” on page 53) and you have viewed your initial results in the
DeltaCue change display viewer (see “DeltaCue Change Display
Viewer” on page 69), you can iterate on those results by changing
various settings and noting the differences in the change display
viewer. If you believe that a particular collection of settings have
broader applicability, you can save those settings in a DeltaCue
settings file (.dqs file) and even distribute that file to other users.
The next time you run the DeltaCue wizard those settings are
available for use.
Iteration Dialog
The DeltaCue Iteration dialog, shown below, is opened from within
the DeltaCue change display viewer by selecting the Iterate button
.
There are three tabs on this dialog; Change Algorithm, Change
Filters, and Material Filters.
Iteration Dialog
81
You must first load a project file into the change display viewer in
order to open the Iteration dialog. The process reads the existing
iterations within the project file and makes them available to you
through the base iteration dropdown menu.
The first step in creating a new iteration is to select a base iteration
from the pull down menu at the top in the section that is common to
all tabs. The last iteration in the project file is displayed by default.
As you select different base iterations, the settings displayed change
to reflect the settings in effect during that iteration. This way you can
review what settings were used for a particular iteration and then
use a given set as the starting point for a new iteration.
The next step is to modify the processing settings. Note that the
image subset and normalization settings are not shown since they
are common to all iterations.
82
Iteration Dialog
To create a new iteration, select the tab that contains the process
setting you wish to change. The new iteration can contain several
changes from the previous one. The parameters in each tab are
discussed in subsequent sections.
Once you have specified a set of processing parameters and have
entered a new output image name, the OK and Record Iteration
buttons become active. You may use the Record Iteration button
to add several new iterations to the project file. Each new iteration
must have a unique output image name. When the OK button is
selected, the recorded iterations are run in sequence. The last
iteration is brought up in the viewer once all iteration processing is
complete. If you simply select OK without recording any iterations,
a single iteration is run using the currently selected processing
parameters.
Note that each iteration may be processed over its own AOI. To
specify an AOI, check the box labeled “Use AOI File” to activate the
AOI file selection box. Then select an existing AOI that has been
saved to a file. You must use an AOI that has been saved to a file,
rather than a temporary AOI in a viewer. If you have an AOI in a
viewer, first save it to a file.
Change Algorithms
Iteration Dialog
The processing settings are identical to those available in the
DeltaCue wizard. You can select the change algorithm from the
available algorithms which are:
•
Tasseled Cap (TC) Transform Greenness difference
•
Tasseled Cap (TC) Transform Soil difference
83
•
Overall Magnitude difference
•
Primary Color Difference (red, green, blue diff)
•
Single-Band Difference
•
Band-Slope Difference
Each difference is a relative difference of the form
D=
(T2 − T1 ) + (T2 − T1 )
T1
T2
where T1 and T2 are the change quantities at Time 1 and 2
respectively. For example, if the Tasseled Cap Greenness
difference is selected, the change quantity is the greenness band
of the Tasseled Cap transform output.
The Tasseled Cap difference algorithms are only available for
sensors that have established Tasseled Cap transform
coefficients. When you first created the project, you specified
the sensor type (see “DeltaCue Wizard Interface” on page 53).
If you selected OTHER as the sensor type, the Tasseled Cap
algorithms are not enabled.
In conjunction with the change algorithm you also specify a
change threshold percentage and whether you want to
interactively review and alter the change thresholds. The
threshold percentage is a +/- threshold that is applied to the
change detection output to eliminate the center portion of the
change histogram. This value is a difference ratio expressed as
a percentage. If you do not interactively review the thresholds,
the process will eliminate any change with values between the
positive and negative threshold values.
If you elect to review the change detection thresholds, when the
processing for this iteration is run, the process will display a viewer
with the thresholded change as an overlay and also display a plot of
the change histogram. You can adjust a set of four thresholds, two
upper and two lower, using interactive controls. Only change
between the two upper and two lower thresholds is retained.
For more information about setting thresholds interactively, see
"Setting Interactive Change Thresholds" on page 65.
84
Iteration Dialog
If you selected Interactive Thresholds in the previous iteration and
you want to reuse that set of four thresholds, select the option to use
a previous iteration’s thresholds.
In that case the process will use the threshold values from the
iteration selected in the pull-down tool.
Change Filters
Three types of changes filters are available – spectral segmentation,
misregistration, and spatial filtering. Spectral segmentation groups
change pixels according to spectrally similar before and after classes
so that you can subsequently filter on those classes. The
misregistration filter attempts to filter out unwanted change that is
merely due to errors in image pair co-registration. The spatial filter
eliminates change based on contiguous size and shape.
Spectral Segmentation
Spectral segmentation is a process that is applied to the Time 1 and
Time 2 images to classify the change pixels into spectrally similar
classes. In the change display viewer, you can then interactively
filter out change pixels based on their before or after spectral class.
If you want a simpler output that merely indicates positive and
negative changes, uncheck the Spectral Segmentation option and
the process will skip classifying the Time 1 and Time 2 change pixels.
This saves processing time, but the resulting change image only
indicates whether the brightness change was positive or negative,
and not the spectral class before and after.
Misregistration Filter
Even though two images have been co-registered some minor
misregistration may remain, typically on local scales. Such pixel
misregistrations can cause apparent change differences simply
because the algorithm was applied to pixels that were not
spatially coincident. The misregistration filter checks the local
neighborhood surrounding each potential change pixel. A local
search window is used to constrain the search for misregistered
pixels. The size of the search window is specified in the Iteration
dialog, as shown below. The filter will search within this window
to see if another pixel in that window would satisfy the definition
of no significant change. If a match is found, then that pixel is a
candidate for being a misregistered pixel.
Iteration Dialog
85
Since local misregistrations typically resemble a shift in pixels, the
process tries to validate the misregistration candidate by examining
nearby neighbors to see if they follow the same pattern of no change.
The number of nearby neighbors considered is controlled by the
search window size. This parameter may be reduced to reduce run
time if needed.
Spatial Filter
The spatial filter identifies contiguous blobs of detected change
by detecting the contour of the blob. Internal holes are
considered part of the total change area. A contiguous blob is a
set of pixels that are connected by at least one neighbor in any
of eight directions. Two change areas are connected if they
share at least one neighbor in common.
The spatial filter eliminates contiguous blobs of detected change
based on their spatial size, either in terms of area or extent, as
defined below. The spatial filter identifies change blobs based on
the binary change mask provided as input. Typically this mask
is produced by the initial thresholding program. Spatial filtering
can be performed in terms of area or extent. Area is the total
area of the contiguous blob, expressed in terms of square
meters. If you select area as the spatial filter type, you must
specify a minimum and maximum blob area.
Once the spatial filter process has detected the contour of a
contiguous change area, it computes several geometric properties
based on the contour. The geometric properties considered are area,
perimeter, major and minor axis length, compactness, and
elongation
See “Spatial Filtering” on page 61.
Area and major/minor axis length have units associated with
them. You can select the units from a dropdown menu. For
example, if the units are meters, the area is in square meters
and the lengths are in meters. Compactness and elongation are
dimensionless quantities.
86
Iteration Dialog
To filter on a given shape property, check the check box
associated with that property and specify the minimum and
maximum values to keep. Note that the range is inclusive. That
means that both the minimum value and maximum values are
included and their corresponding shapes retained. Thus, if you
specify a minimum and maximum area of 10 pixels, only shapes
with exactly 10 pixels are retained. If the maximum area is set
to 11 pixels, then shapes with both 10 and 11 pixels are
retained. However, if the minimum area is set to 10.1 pixels,
then only shapes with 11 pixels are retained.
Once you have adjusted the change algorithm and change filter
settings for the new iteration, you must enter a change detection
output file name before the iteration can be run. The dialog Ok
button becomes enabled once you specify a file name. Use the file
browser button to specify the directory path for the output file name.
When you select the Ok button in the Iteration dialog, the
process updates the existing project file (.dqw file) to include the
new iteration settings and it launches the dQrunprocess
program to execute the new iteration. If the settings for the
change algorithm have not changed, then the intermediate files
related to change detection are not recreated, but the existing
files are used. This saves time by not having to re-run processes
for existing quantities that will not change in the new iteration.
The process only creates new quantities when the related
settings are different from the base iteration or when settings
upstream in the processing chain have changed. The process
tries to minimize the amount of re-processing that must be
performed.
Material Filters
The third tab on the Iteration dialog is used to specify Material
Filtering. Material Filtering is only available for sensors that have
Tasseled Cap coefficients derived for them. If you selected one of the
sensor types to be OTHER in the "Project Selection Dialog" on page
54, then the Material Filtering section is grayed out since Tasseled
Cap coefficients are not available for that sensor pair.
Material filtering allows you to exclude selected categories of
materials from the change detection result. For example, by
selecting the check box for Vegetation in Time 1, you are electing to
exclude any change that includes vegetation in Time 1. Changes that
involve vegetation in Time 2 are not excluded unless you explicitly
check that box.
Iteration Dialog
87
The material categorization process is intentionally
conservative. The process generally does not exclude change
pixels that include mixtures of material categories since those
may be of interest. The material filter only applies to those pixels
that clearly fit the category. Other change pixels will have to
filtered out by other means.
Once you select a Material Filter category by checking the check box
next to the category in either Time 1 or Time 2, the parameters
associated with that material filter become enabled. The default
values generally apply since they are conservative.
Refer to “DeltaCue Material Filtering” on page 105 for a
description of the Material Filtering process.
Saving Settings
You can save the current iteration settings at any time using the
Iteration dialog Save button. In that case, the following Save
Settings dialog is displayed.
You must enter a descriptive name for the settings being saved and
a short description of what the settings are or what they apply to.
The Settings Name will appear in the DeltaCue wizard parameter
settings the next time you run the wizard.
Note that the settings saved are those currently visible in the
Iteration dialog, not the base iteration settings. You can change
the settings and save them without actually running the
iteration.
When you save settings, the current settings are stored in a
DeltaCue settings file (.dqs file) which is a special form of project file.
Your setting file is stored in your temporary directory since you
automatically have write permission to that directory. The DeltaCue
wizard knows to look in that directory for settings files.
88
Iteration Dialog
Settings files are automatically named dq<settings name>.dqs
where <settings name> was the name string that you entered
when saving the settings file. If you want to delete previously
saved settings from the DeltaCue wizard settings dropdown
menu, delete the corresponding settings file from your
temporary directory. Simply locate the .dqs file with the
corresponding name, delete it, and re-run the wizard.
Some settings files are distributed with the software, such as the
program defaults. These settings files are installed with the software
and are located in the directory <IMAGINE_HOME>/etc/DeltaCue. If
you wish to distribute your settings files to other users, they should
place them in this directory. Certain administrative privileges may be
required to access this directory. Check with your Systems
Administrator.
Iteration Strategy
DeltaCue iterations can be used to try different filters and filter
settings, different change thresholds, and different algorithms. The
number of iterations can quickly grow and a strategy is useful for
constructing iterations and managing them. Also, in order to
minimize run time for each iteration, it is best to make changes in
parameter settings as far down the processing chain as possible so
that the process will not re-run intermediate processes unnecessarily.
The DeltaCue processing chain consists of:
•
Compute change quantities
•
Difference change quantities
•
Threshold change output (creates mask image)
•
Apply Misregistration filter to change output (creates mask
image)
•
Apply Material filter to change output (creates mask image)
•
Apply Spatial filter to change output (creates mask image)
•
Apply mask images to create final change detection output
•
Perform Spectral Segmentation of remaining change pixels
The basic iteration strategy recommended here is as follows:
•
Iteration Strategy
make settings changes that affect processes at the start of the
chain first and then work your way down the chain
89
•
save the use of filters for later iterations so that you begin with
the maximum amount of change available and then reduce that
amount
For example, your first iteration should specify a change algorithm,
such as Tasseled Cap Greenness difference, with no change filters
applied. This choice of algorithm affects the change quantities and
their difference. Depending on your choice of threshold percentage,
you should see a relatively large amount of change. You might next
iterate using different change thresholds to reduce the amount of
unwanted change, such as that due to clouds and cloud shadows.
Once you have a reasonable set of change thresholds, you might
start applying change filters to further reduce the final amount of
change presented to you. If those results are not sufficient, then you
could try a different change algorithm and work your way back down
the chain. Making changes later in the chain saves processing time
since early changes do not propagate down through the chain.
You always want to begin iterating with a change detection
algorithm that is appropriate for your application. For example,
if you are interested in changes to vegetation, you would want
to consider using the Tasseled Cap Greenness difference. If you
are interested in soil changes, use the Tasseled Cap Soil
difference or the magnitude difference algorithm. Carefully
inspect this initial iteration to see that the algorithm is capturing
the sort of change you are interested in. If not, try a different
algorithm. You want your initial results to contain lots of
appropriate change and if there is some unwanted change, you
will try to filter it out in subsequent iterations. A strict change
threshold that removes most all unwanted change may also
remove some more subtle varieties of the change in which you
are interested. With use you will begin to get a feel for how
conservative or liberal you should be when setting a change
threshold for a given change phenomenon of interest.
90
Iteration Strategy
DeltaCue Site Monitoring Mode
The Site Monitoring mode within DeltaCue software is intended for
those applications where you wish to identify changes at a specific
site rather than search for change over broad areas. In this mode,
you identify an Area of Interest using the IMAGINE AOI tools and the
software subsets the image pair down to this area and displays
change in a custom site-monitoring change display viewer. The
viewer provides several change detection visualization tools and
enhancements to help you quickly interpret changes at the site.
A Site Monitoring process assists you in automatically subsetting the
Time 1 and Time 2 images down to the area of interest and then
computes a difference image over the subset. It also layerstacks the
two images into a multitemporal image with the Time 1 image in
layers 1 through N and the Time 2 image in layers N+1 to 2N. If the
Tasseled Cap coefficients are available for the images, the difference
image can be based on the brightness, greenness, and wetness
bands of the Tasseled Cap transform. This allows you to visualize
change in a more physically meaningful way.
Two visualization modes are available with DeltaCue Site Monitoring
software – Material View and Multitemporal View. Material View is
only available for images with Tasseled Cap coefficients and presents
change in terms of the brightness, greenness, and wetness
difference following a Tasseled Cap transform of each image.
Multitemporal View simply computes the difference on the original
bands. You select the Time 1/Time 2 band difference and
corresponding layerstack to see changes in that band.
Site Monitoring
Process
The DeltaCue Site Monitoring process is a sequence of processing
steps based on inputs that you provide. Click Site Monitoring from
the DeltaCue main menu to open the Site Monitoring dialog, shown
on the next page.
Site Monitoring Process
91
The Site Monitoring process stores intermediate results in a project
workspace directory and captures the inputs in a project file (.dqm
file).
Create a New Project
Begin by creating a new project. If you have an existing project, you
can elect to open that project instead and the Site Monitoring viewer
is opened without re-running the process.
To create a new Site Monitoring project:
1. Click Create a New Project radio button.
2. Enter a project file name in Project File Name field. If needed, click
the file browse icon to change directories.
The process will automatically create a workspace directory in the
same directory as your project file and with the same base name.
For example, if you specify a project file called C:\test\site.dqm, the
process will create a workspace directory called site in C:\test. The
intermediate difference and layerstack images are placed in this
workspace directory. When you reopen a project, you need only
specify the project file. The software automatically finds the files in
the workspace.
3. Enter the Time 1 and Time 2 image names in the respective fields.
These images must be co-registered to each other using a process
such as IMAGINE AutoSync, but they need not be cropped to a
common area.
4. Select the sensor type for each image by clicking in the Sensor
Time 1 and Sensor Time 2 dropdown menus.
If your sensor does not appear, select Other as the sensor type. If
you select Other, Material View is disabled since it depends on
sensors that have established Tasseled Cap coefficients.
92
Site Monitoring Process
5. Enter the AOI File Name.
If you do not have an existing AOI that encompasses your site of
interest, open an IMAGINE viewer and create an AOI around the site.
Then save that AOI to a file. You must save your AOI to a file since
the Site Monitoring software does not use temporary AOIs in the
viewer by design. If you want to run the site monitoring process over
an entire image, simply draw an AOI that encompasses the entire
image.
6. Click OK. A project file (.dqm file) is created and a number of
separate processes are spawned to create intermediate outputs.
Each separate process will create its own progress dialog. When all
of the processes in the processing chain have completed, the
dQrunsitemonitor progress dialog indicates completion.
Use an Existing Project
If you elect to use an existing project, the Site Monitoring dialog
changes to the one shown here.
Click the file browse icon to open an existing project file and select
OK. The Site Monitoring Viewer is launched with the selected project
as input.
Site Monitoring
Viewer
The DeltaCue Site Monitoring Viewer is similar to the regular
DeltaCue change display viewer. They share a number features and
controls in common so that once you become familiar with one, the
other viewer will also be familiar.
The Site Monitoring Viewer is configured in one of two modes
depending on your selection of Material View or Multitemporal View.
The main difference between the two modes is the type of change
image loaded and the viewer control at the top.
Site Monitoring Viewer
93
Material View
The Material View is available for sensors that have Tasseled Cap
coefficients. In this view, the change difference is based on the first
three Tasseled Cap transform layers which are Brightness,
Greenness, and Wetness. In addition to differencing these layers,
the Site Monitoring viewer also provides a view of a multitemporal
layerstack of these layers.
An example of the Material View mode is shown here:
The left side of the viewer contains the pair of AOI-subsetted images
with Time 1 on top and Time 2 on the bottom. The right side contains
the layerstack image on the top and the difference image on the
bottom. As in the change display viewer the two windows are linked
so that panning in one window will cause the other window to pan
similarly. The two windows’ zoom levels can be synched using the
zoom synching tool and the change magnifier behaves as in the
change display viewer.
If you wish to view the lower image in the right-hand window, click
the Change Background icon
to turn off the upper layer in the
window so that the underlying layer is visible.
The Time 1/Time 2 layerstack of the Tasseled Cap transform image,
which is the top-most image in the right-hand viewer, provides a
simultaneous view of two dates of one of the tasseled cap
components. Select which quantity to view using the radio buttons
above the right-hand window. Select only one quantity at a time for
display.
94
Site Monitoring Viewer
You can also select Regular or Invert to reverse the order of the
Time 1 and Time 2 images in the layerstack display.
The various quantities’ associated color channels or color guns are
roughly related to the components (red to soil, green to greenness
and blue to wetness).
For example, Brightness in Time 2 (Regular mode) is displayed in the
red channel and Brightness in Time 1 is displayed in the blue and
green channels. Increases in Brightness will appear as reddish pixels
while decreases will appear cyan. Greenness in Time 2 is in the green
color channel, while Time 1 is in red and blue, therefore increases
will appear as greenish pixels, decreases in magenta.
Similarly, Wetness in Time 2 is in the blue channel and increases in
its value will appear as bluish pixels, decreases as yellow pixels. This
leads to an intuitive interpretation scheme of “if something is bright
red, green or blue, that indicates an increase in the associated
physical variable (soil, vegetation or water) between the two dates.
The tables below summarize the color schemes for the layerstack
image.
Quantity
Regular
RGB
Brightness
Red = increase in Time 2
T2, T1, T1
Cyan = decrease in Time 2
Greenness
Green = increase in Time 2
T1, T2, T1
Magenta = decrease in Time 2
Wetness
Blue = increase in Time 2
T1, T1, T2
Yellow = decrease in Time 2
Quantity
Invert
RGB
Brightness
Red = decrease in Time 2
T1, T2, T2
Cyan = increase in Time 2
Greenness
Green = decrease in Time 2
T2, T1, T2
Magenta = increase in Time 2
Wetness
Blue = decrease in Time 2
T2, T2, T1
Yellow = increase in Time 2
Site Monitoring Viewer
95
The layerstack image can provide you with a quick and intuitive
depiction of change in a physical phenomenon across an area of
interest.
For instance when viewing the greenness image, shades of green
give an indication of how much vegetation has grown between the
two dates, while Invert will display in green the loss or dieback of
features related to vegetation. For some users the inverted display
scheme may be more familiar. A mnemonic device sometimes used
for image interpreters with the inverted layerstack image is “blue is
new, red has fled” – meaning things that are displayed as red are
from the Time 1 image and blue pixels indicate a greater response in
the Time 2 image.
Under the layerstack image is the difference image of each of the
three Tasseled Cap components. These are the images that would
normally be thresholded for significant change using the interactive
threshold tool or percent change option.
See "Change Threshold" on page 30 and "Change Detection
Dialog" on page 57.
In the site monitoring mode these images are not thresholded but
are displayed in a manner to highlight gradations of change in either
the positive or negative direction (pixels that got brighter or darker
between the two dates respectively) for each of the Tasseled Cap
components.
The difference image contains three layers which are mapped to the
red, green, and blue guns on the display. As in the layerstack layer,
Tasseled Cap brightness is mapped to the red channel, the difference
in greenness is mapped to the green channel, and the difference in
wetness is mapped to the blue channel. All three colors can be
displayed simultaneously. You can control which quantities are
shown using checkboxes in the control panel.
In addition to which quantity is shown in the difference layer, you
can also control whether positive or negative changes from Time 1
to Time 2 are displayed. The Positive and Negative radio buttons
control the red, green, and blue lookup tables so that only positive
or only negative changes are displayed in the lower difference
image.
96
Site Monitoring Viewer
For example, if you want to see increases in vegetation (greenness)
from Time 1 to Time 2, select the Greenness channel and Positive
changes. The right-hand display will change to show only positive
greenness changes (in green). Likewise, if you want to see decreases
in vegetation over time, select the Negative option. The display will
change such that bright green pixels represent large decreases in
greenness over time. The direction (positive/negative) indicates the
type of change and the intensity on the screen indicates the degree
of change.
You can also control the standard deviation stretch. Click in the field
under SD to change the standard deviation stretch for the difference
layer. Select a value between 1 and 6 standard deviations.
The intensity values are clipped at the selected standard deviation
value so that any values greater appear at the same intensity. Small
values are used to enhance subtle differences and larger values are
used to see big changes between the two dates.
The Blend Tool
is used to interactively blend the top layerstack
image into the bottom difference layer. An example is shown here.
Site Monitoring Viewer
97
As you move the slider bar back and forth, the right-side display
blends from the top layerstack layer to the bottom difference layer.
This can be useful for identifying changes of interest or for
associating features in the difference image with features in the
layerstack image.
It is best to return the blend slider bar to 100% before closing
the Blend Tool. This restores the top layer to its original state.
Refer to On-Line Help for more information on the Blend Tool.
The Background Toggle tool
can also be used to flicker between
the two layers on the right side of the viewer.
Multitemporal View
The Multitemporal View is used when you want to examine change
apparent in particular bands or when the Tasseled Cap coefficients
are not available with your sensor data. In Multitemporal View, the
right side of the display contains the layerstack of a selected band on
top and the difference image of that band on the bottom layer. As
this view displays only one band at a time, the lower difference
image is a grayscale.
The upper layerstack image is a Time 1/Time 2 layerstack of the
selected band. The red channel is always used to display the selected
band at Time 2, so increases always appear reddish and decreases
always appear cyan in color, unless the Invert radio button is
selected, whereupon the color scheme will reverse. An example of
the Multitemporal view Site Monitoring Viewer with the top
layerstack image turned off is shown below.
98
Site Monitoring Viewer
The difference image is displayed as a grayscale image where
brighter pixels indicate greater increases in that band from Time 1 to
Time 2.
If you wish to view the lower image in the right-hand window, the
Change Background tool
can be used in the Site Monitoring
viewer to turn off the upper layer in the window so that the
underlying layer is visible.
The viewer controls for this mode allow you to specify the band of
interest and to control the layer stack and difference image display
options.
The Band menu is used to specify which band is used for both the
upper layer stack and the lower difference image.
Under the Layer Stack text are the radio buttons to control the
display of the layerstack image. Regular is time 2 in red and time 1
in blue/green. Invert is time 1 in red and time 2 in blue/green.
Under the Difference Image text are the radio buttons to control
whether to display positive (Pos) or negative (Neg) changes.
Site Monitoring Viewer
99
The SD Stretch menu allows you to select a standard deviation
stretch for the difference layer. Values between 1 and 6 standard
deviations are available. The intensity values are clipped at the
selected standard deviation value so that any values greater appear
at the same intensity. Small values are used to enhance subtle
differences and larger values are used to see big changes between
the two dates.
The Blend Tool
is used to interactively blend the top layerstack
image into the bottom difference layer.
As you move the slider bar back and forth, the right-side display
blends from the top difference layer to the bottom layerstack layer.
This can be useful for identifying changes of interest or for
associating features in the difference image with features in the
layerstack image.
It is best to return the blend slider bar to 100% before closing
the Blend Tool. This restores the top layer to its original state.
Refer to On-Line Help for more information on the Blend Tool.
100
Site Monitoring Viewer
DeltaCue Workspaces
When you use the DeltaCue wizard to process an image pair, the
software automatically creates a project file and corresponding
workspace directory. The workspace directory name is derived from
the base project name and is located in the same file path as the
project file. So if your project file name is test.dqw, a workspace
directory called test is created in the directory where the project file
is located.
The site monitoring mode works under the same premise,
creating a workspace folder of the same name as the .dqm file
where all the necessary files for the site monitoring mode are
stored. If you create a .dqm file with the same name as an
existing .dqw file, DeltaCue will place all the site monitoring files
in the same folder where the broad area search mode-related
files are stored. Both the .dqw and .dqm files will still be
readable by the change display viewer and Site Monitoring
Viewer, respectively.
The workspace directory will contain a number of intermediate
processing output files. As you iterate with DeltaCue software,
the process attempts to reuse existing outputs files whenever
possible to save processing time. The software looks to the
workspace directory to see if certain output files already exist
before recreating them. If a process is interrupted or exits
abnormally, it is possible that it may leave a corrupted version
of an intermediate output file in the workspace directory. In that
case you should manually delete any files that appear to be
causing problems.
The DeltaCue processes use a naming convention for intermediate
files in any workspace directory, so it is easy to determine which files
have been produced and their purpose. The following table describes
the naming conventions for the files associated with the Broad Area
Search mode, that is, files referenced by the .dqw file.
101
File Name
Description
subset1.img
Common area subset images for Time 1 and
Time 2.
subset2.img
norm2.img
Normalized Time 2 image.
class1.img
class2.img
Isodata classifications of Time 1 and 2 which
will exist if during normalization you indicated
that there were clouds present.
ma1.img
Magnitude image for Time 1 and Time 2.
ma2.img
tc1.img
tc2.img
Tasseled Cap transform images for Time 1 and
Time 2.
Each image contains a layer for greenness and
one for soil.
sb1-n.img
sb2-n.img
bs1-n.img
bs2-n.img
re1-n.img
re2-n.img
gr1-n.img
gr2-n.img
bl1-n.img
bl2-n.img
mag-n.img
Single band image for Time 1 and 2 for
iteration n.
Band slope image for Time 1 and 2 for iteration
n.
Red color image for Time 1 and 2 for iteration
n.
Green color image for Time 1 and 2 for
iteration n.
Blue color image for Time 1 and 2 for iteration
n.
Magnitude difference image for iteration n.
Note that a given image may be used for
several iterations.
tcg-n.img
Tasseled cap greenness difference image for
iteration n.
Note that a given image may be used for
several iterations.
tcs-n.img
Tasseled cap soil difference image for iteration
n.
Note that a given image may be used for
several iterations.
sbd-n.img
102
Single band difference image for iteration n.
bsd-n.img
Band slope difference image for iteration n.
red-n.img
Red color difference image for iteration n.
grn-n.img
Green color difference image for iteration n.
blu-n.img
Blue color difference image for iteration n.
XXXthresholds-n.txt
Text file of change thresholds for iteration n
and change algorithm XXX where XXX = tcg,
tcs, mag, etc.
XXXmask-n.img
Initial change threshold mask for iteration n
and change algorithm XXX where XXX = tcg,
tcs, mag, etc.
XXXmisregmask-n.img
Misregistration filter mask for iteration n and
change algorithm XXX where XXX = tcg, tcs,
mag, etc.
XXXmaterialmask-n.img
Material filter mask for iteration n and change
algorithm XXX where XXX = tcg, tcs, mag, etc.
XXXspatialmask-n.img
Spatial filter mask for iteration n and change
algorithm XXX where XXX = tcg, tcs, mag, etc.
XXXspatialmask-n.txt
Text file of spatial filter parameters for all blobs
in change mask.
Time1ClassImage.img
Spectral segmentation of Time 1 and Time 2
change pixels.
Time2ClassImage.img
The DeltaCue project file (.dqw file) is an XML-based text file that is
used to record the initial project settings and those used during all
iterations. The project file contains XML tags for the various files and
processing parameters used by the process.
A list of the XML tags used in the project file is shown below.
103
<WORKSPACE>
<NAME></NAME>
<WORKSPACEPATH></WORKSPACEPATH>
<COMMON>
<TIME1_INPUT></TIME1_INPUT>
<TIME2_INPUT></TIME2_INPUT>
<SUBSET_IMAGES SUBSET="Y/N">
<TIME1_SUBSET></TIME1_SUBSET>
<TIME2_SUBSET></TIME2_SUBSET>
</SUBSET_IMAGES>
<NORM_IMAGES NORMALIZE="Y/N">
<TIME1_CLOUDS>Y/N</TIME1_CLOUDS>
<TIME2_CLOUDS>Y/N</TIME2_CLOUDS>
<NORM_FILE></NORM_FILE>
</NORM_IMAGES>
</COMMON>
<ITERATIONS>
<ITERATION NUMBER="1">
<OUTPUT_FILE></OUTPUT_FILE>
<CHANGE_DETECTION CHANGE_TYPE="XXX">
<CHANGE_FILE></CHANGE_FILE>
<TIME1_TRANSFORM></TIME1_TRANSFORM>
<TIME2_TRANSFORM></TIME2_TRANSFORM>
<THRESHOLDS THRESH_APPLY="Y/N">
<PERCENTAGE></PERCENTAGE>
<THRESHOLD_FILE></THRESHOLD_FILE>
<THRESHOLD_MASK></THRESHOLD_MASK>
</THRESHOLDS>
</CHANGE_DETECTION>
<MISREGISTRATION MISREG_APPLY="Y/N">
<MISREG_MASK></MISREG_MASK>
<MISREG_WINDOWSIZE></MISREG_WINDOWSIZE>
</MISREGISTRATION>
<SPATIAL_FILTER SPATIAL_APPLY="Y/N">
<SPATIAL_MASK></SPATIAL_MASK>
<SPATIAL_MINAREA></SPATIAL_MINAREA>
<SPATIAL_MAXAREA></SPATIAL_MAXAREA>
</SPATIAL_FILTER>
</ITERATION>
</ITERATIONS>
</WORKSPACE>
104
DeltaCue Material Filtering
The material filtering process in DeltaCue software is based on the
Tasseled Cap transformation which transforms a source image from
its wavelength-based band spectrum to a space with more physical
meaning. The Tasseled Cap transformation is derived via the GramSchmidt orthogonalization method that produces a linear
transformation which aligns the first basis vector with the mean soil
brightness followed by orthogonal basis vectors along the greenness
direction and the wetness direction and so on. The projection of the
pixel spectra onto the plane defined by the first basis vector (soil
brightness) and the second basis vector (greenness) are naturally
clustered into obvious groups and can be separated according to
angles measured from the negative direction of the greenness axis.
Material Filter
Parameters
In the case of a Tasseled Cap transformation derived using the
Gram-Schmidt orthogonalization method, the theoretical angle of a
non-vegetation material (including pure soil) is 90 degrees since it
should align with the first basis vector.
Pixels representing vegetation have an angle higher than nonvegetation while water and shadow and bright (but flat in source
spectrum) materials have angles lower than that of non-vegetation.
There is a fuzzy zone between the vegetation and non-vegetation
and another fuzzy zone between the non-vegetation and water (and
shadow and bright but flat spectrum material).
For Quickbird II and IKONOS sensors, the vegetation/nonvegetation boundary is about 100 degrees and the nonvegetation/water-shadows boundary is 80 degrees.
For Landsat sensors, those two angles are 86 and 82 respectively.
The angle of 86 is less than 90 is because the non-vegetation axis is
not aligned with the first principal component when the Tasseled Cap
Transformation coefficients were derived. The bright but flat
materials like clouds are within the same angle zone as that of
water/shadow and another parameter – the soil brightness can be
used to separate those two. That value is in the unit of DN and is 500
for Quickbird II and IKONOS.
Material Filter Parameters
105
106
Material Filter Parameters
Index
A
absolute difference 12, 13
accompanying image 13
acreage estimates 23
actual change 13
after class material 21
agricultural
change 31
lands 10
Analyzing Change 22
annotation layers 73, 74
AOI
file 64, 83
apparent
changes 7
image change 13
Arrange Layers 72
control 72
asphalt parking lot 26
associated phenomena 24
atmospheric
conditions 26
corrections combined 21
differences 7
haze 11
automated cropping procedure 24
automated processing steps 7, 8
automatic swipe mode 77
B
Background Toggle tool 98
band
basis 25
combination 80
comparison 12
band-slope computed 29
Band-Slope Difference 25, 84
bare
soil 21, 49
base
project name 101
base iteration 82, 87
before class material 21
Blend Tool 97, 100
blue
diff 84
bounding thresholds 16
bright dry sand 26
brightness
change 22, 24, 61, 85
Index
magnitude 25
value 13, 26, 30
values 10, 11
Brightness control 76
building construction 26
C
careful registration 19
change
blob 17
class 22
classification techniques 12
detected 11, 19
difference 36
display viewer 5, 6, 9, 45, 47
magnitude 7
mask 86, 103
output 7
overlay 47
phenomena 9
phenomenon 17, 90
quantity 84
related 13, 18
results 22, 32
searches 23
simultaneously 70
value 12, 14
change algorithm 58, 87, 90, 103
Change Algorithms 25, 83
Change Analysis 45
Change Background 79, 99
Change Background tool 99
Change Detection 1, 3, 5, 6, 7, 8, 9, 10, 12,
15, 17, 21, 22, 23, 25, 29, 35, 36, 53, 57,
69, 74, 81, 87, 90
Basics 9
Methods 12
Viewer 53, 81
change detection
algorithm 8, 15, 22, 29, 90
analysis 3, 69
method 17
methodology 8, 23
output 5, 87
process 3, 7, 36
set 74
strategy 23
studies 10
task 23
Change detection dialog 57
Change Display
Viewer 3, 21, 35, 69, 81
change display viewer 53, 69, 81, 85, 101
107
change filter 87
settings 87
Change Filtering 30
Change Filters 8, 60, 85, 90
Change filters 60
Change filters dialog 60
change histogram 47
change image 12, 13, 14, 15, 16, 18, 19,
22, 24, 30, 85
Change Magnifier 71, 79
change region 17, 18, 30
blob 17
Change Results Viewing 32
Change Threshold 12, 13, 14, 15, 30, 33,
84, 90, 103
change threshold
percentage 84
Change thresholds 14, 15, 84, 90, 103
changed block 23
characteristic dimensions 17
circular region 17
classed range 22
classes
displayed 80
represent 20
Clear Viewer 74
cloud shadows 7, 16, 25, 90
clouds present 102
common
coverage 24
neighbor 30
Compactness Values 18
computing relative difference 25
construction activity 40
contiguous
blob 30, 61, 86
change 86
size 85
continuous
blob 31
coordinate reference system 11
co-registered image 7, 11
data 7
pair 11
cosine values 59
create new
AOI 72
Cursor Tool 75
D
dark disturbed soil 49, 50, 51
data
mean centered 14
108
source 24
types 23
deciduous trees 10
decision support 22
delineate polygons 23
detect
new buildings 17
detecting
change 3, 12, 23
man-made objects 8
detection
process 3, 7, 11, 23, 36
sizes 17
difference
operation 8
ratio expressed 84
difference image 15, 24, 102, 103
different
crop types 10
times 10
differential haze 25
direct technique 12
discriminate various shapes 18
disregard contiguous groupings 17
distribution tailing 14
dqm file 92, 93, 101
dQrunprocess program 65, 87
dqs file 32, 69, 81, 88, 89
dqw file 32, 54, 65, 74, 87, 101, 103
E
earth-based spatial framework 11
eliminate
changes 8
detected change 10
fields based 52
elongation value 17
example data 1
Excessively conservative thresholds 23
existing
tools 22
Existing project 67
extreme
vegetation-related 24
Extreme changes 15, 26, 30
extreme tails 16
F
fallow periods 10
false change detections 16
File
menu 72
Open 74
Index
file
name 38, 101, 102
path 101
filter change 8, 17
based 8
regions 17
filter settings 23, 87
filtering procedures 22
Filtering Unwanted Change 16
fire scars 26
first
iteration 47
fitting criteria indicative 10
flicker control 70, 77
Flicker Tool 76, 77
forest
loss 23
removal 26
Four thresholds 66
G
Geometric
compactness 17
properties 8, 30, 62, 86
geometric
characteristics 17
distortions 10
moments 17
properties based 86
properties considered 86
transformations 10
GIS layer 23
Gram-Schmidt orthogonalization method 105
grayscale 12
image 12
Greenness
Differencing 25
greenness
band 84
change 22
difference 26
difference image 102
direction 105
H
highlighted
change areas 22
pixels 22
histogram
plot 47
window 43
homogeneous haze 25
Index
I
icon panel 3
illumination changes related 10
illumination conditions 11, 57
illumination differences 11, 13
Image
co-registration 24, 36
registration 9, 10, 24
image
cropping 55
data 5, 7, 8, 79
environments 23
grids 11
misregistration 7
normalization 24, 25
pair 1, 9, 10, 11, 12, 15, 18, 24, 85, 101
pair co-registration 85
pixel 12
processing 7, 9, 22, 23
processing techniques 23
relative 11
statistics 14
transformation 13, 24
image acquisition 10, 23
Image Analyst 23, 32, 33
Image cropping dialog 55
Image Scientist 8, 23, 30, 32
image-to-image normalization 25
Image-to-image registration 9
impervious surface 23
impervious surface growth 23
industrial facilities 22
initial
image 12
project settings 103
threshold 60
Initial Processing 36
Initial Wizard Usage 53
input
file names 9
images 42
Inquire Cursor 72
Insignificant change 7, 29
intense vegetation changes 24
Interactive Thresholds 59, 60, 65, 85
intermediate
files 101
output file 101
Internal holes 86
interpolation process 11
interpreted
context 22
information class 21
invalid change 10
109
Isodata classifications 102
iteration
based 65
capability 35, 69
menu 46
Iteration Dialog 76, 81
Iteration Selection 76
Iteration Strategy 89
L
landcover
classes 20
material 45
materials 7, 8, 80
types 20, 21
Landsat TM 29
imagery 29
latitude/longitude coordinates 11
Layer swipe 70, 77
leaf-off condition 10
leaf-on state 10
linear feature 18, 19
linear transformation 25, 26
equations 26
local
misregistration 19
scales 85
longer side 17
lower
thresholds 66
lower bounding thresholds 16
lush active vegetation 51
lush vegetation 50
M
Magnification levels 79
magnified view 79
magnifier button 71
Magnifier Properties 79
magnifier window 79
Magnitude
changes 45
image 102
Magnitude Difference 24, 25, 46, 48, 90, 102
algorithm 25, 46, 90
magnitude difference
approach 24
image 102
Magnitude Differencing 25, 26
magnitude differencing process 26
magnitude value 26
main
menu 3
110
major
axis 17, 18, 62
dimension 17
principal axis 17
Major Axis Length 62
master image 11
material
categories 42
categorization process 88
change 6
type 22
Material Filter 31, 33, 88, 103
category 88
Material filter dialog 63
Material Filtering 21, 87, 105
section 87
Material filtering 63
Material Filters 87
Material View 91, 94
mathematical center point 17
maximum
blob 86
threshold 7
values 87
Measure Tool 72, 78
measurement tool 78
minimum threshold 7
minimum value 87
minor
axes 17
axis 18, 62, 86
misregistration 85
principal axis 17
Minor Axis Length 62, 86
misregistered
pixel 85
mis-registered pair 18
misregistration candidate 86
misregistration effect 19
Misregistration Filter 7, 33, 46, 52, 61, 85,
103
Misregistration Filtering 18
model integration 23
multispectral
image 13
imagery 26, 29
multitemporal
image 91
multi-temporal imagery 13
Multitemporal View 91, 98
N
naming convention 101
Index
natural spectral variability 13
negative
change 14
values 12
Negative changes 85
negative threshold 15
new
industrial structures 22
residential structures 22
road 7, 17
structures 22
New Project 38
non-change related effects 13
non-vegetative materials 58
normal variability 15
normalization
procedure 25
process 40, 56
Normalization dialog 56
north arrow 78
North Symbol 78
numbered classes 21
O
on-line help 3, 72
open
threshold 24
optimal normalization 25
original
bands 13
image pair 12
imagery 22
images 6, 19
outer
threshold 15
outer bounds 16
output
file name 9
image file 64
Output Image 64
overall
magnitude 13, 24, 84
statistics 25
overlay layers 74
overlays
available 73
P
painted structures 28
Pan Tool 76
panchromatic imagery 26, 29, 58
parameter file 8, 9
parameter file name 9
Index
parameter settings 8, 23
parameter settings file 23
partial detection 17
percentage change 12, 13, 14, 15, 22
value 12, 14
percentage difference image 15
phenomenological-based approach 24
physical landcover types 20
pixel
data 36
grid 10, 11, 36
misregistrations 85
representing 18
spectrum 27
value 10
vector 27
point
coordinates 22
location 23
positive
values 12
Positive change 15, 16
positive change thresholds 15
precisely co-registered 18
precisely registered 10
Preliminary Steps 10
presence/absence events 26
Primary Color
Differences 27
Differencing 25
primary color 8, 25, 27, 29, 58, 84
algorithms 8
principal axes 17
Process Management 32
processing
chain 89
parameters 9, 103
techniques 13, 23
time 12, 101
Processing Strategy 23
production
analyst 8
Program Defaults 38
project
file name 38, 101
workspace directory 6
Project file 32, 38, 88, 101, 103
documents 32
Project selection dialog 54
R
radiometric
correction 12
111
distortions 11
Radiometric Normalization 10, 11, 12, 25
random reflections 7
real change 7, 15
real physical change 13
recorded iterations 65
Rectangle Length 18
reference
image 10
spectra 27
vector 27
relative
change 67
difference 12, 25, 26
formula 12
remove
change classes 21
Reset
button 44
Zoom 74
resolving power 16
restricting detections 23
resulting
change image 12, 18
phenomenological components 13
road pixel 18
rotation axis 62
row/column location 10
S
sample data 1
Saving Settings 88
Scale Symbol 78
scene
feature changes 7
features 21
phenomenology 13
scene elements 13
Scene illumination differences 11
scene-to-scene radiometric difference 12
schematic representation 18, 19
search
mode-related files 101
scenarios 23
window 19, 86
window size 86
Search mode 9, 101
second date 22
second row 77
selected
files 72
Selection Tool 75
sensor
112
acquisition distortions 10
anomalies 11
noise 7
response 25
type 38, 84
Session Output 64
settings
file 23, 32, 33, 69, 88
upstream 87
shape
discriminators 17
filter 33
property 87
shape characteristic 17
shapefile layers 72, 73, 74
Shapefile Output 76
shift-like pattern 61
side zoom 74
significant change 6, 7, 61
single
band 12, 102
Single-Band Difference 25, 84
Site Monitoring 9, 23, 91, 93, 101
Mode 9, 91, 101
Process 91
Viewer 93, 101
site monitoring
applications 23
files 101
perspectives 9
slave
image 10, 11
pixel grid 11
soil
brightness 13, 22, 24, 25, 26, 105
changes 90
conditions 7
disturbances 24
source
imagery 12
spatial
characteristics 8, 20
description 22
filter available 32
filter parameters 103
filter process 86
filter type 86
misregistration 32
properties 8
size 22, 86
Spatial Filter 18, 32, 46, 52, 85, 86, 103
Spatial Filtering 8, 16, 18, 61, 62, 85, 86
specific
application 23
Index
regions 22
specified speed 77
Spectral
Bands 26, 80
characteristics 8, 20, 80
Filtering 20, 21, 80
Segmentation 20, 21, 31, 49, 61, 85,
103
Segmentation option 85
spectral
class 85
classes 21, 80
composition 22
filtering process 21
nature 20
properties 8
segmentation routine 31
segmentation tool 21
signature libraries 21
spectral angle
changed 28
transform 27
speed processing 10
standard deviation 25, 100
spectra 25
stretch 100
standard deviation stretch 97
standardized procedure 9
statistics generation 23
strict change threshold 90
subset images 102
subtle
changes 9
varieties 90
swipe
action 77
control panel 77
tool 77
symmetry 62
Synchronize Zoom 75
unchanged target 13
unsupervised classification process 56
unwanted
after classes 51
before classes 50
change 16, 21, 46, 85, 90
Upper Thresholds 66
T
W
Tasseled Cap
algorithms 84
coefficients 26, 30, 55, 63, 87, 91, 94,
98
components 96
difference 48, 84
differences 26
Differencing 24
transform 91, 102
transformation 13, 22, 26, 105
terrain distortions 10
Index
threshold
mechanism 7
program 47
thresholded change 18, 43, 66
image 18, 66
thresholding process 16, 17
thumbwheel control 77
tide levels 10
Tools menu 73
total change 86
transformed cosine value 28
Transition classes 80
Triangle controls 43
two-dimensional pixel space 17
U
V
Various Rectangular Shapes 18
vegetated field 18
vegetation 97
change 31
vegetation-related change 31
vegetative growth 49
vehicle
traffic 10
Viewer
Controls 70, 74, 77
Features 69
Menus 72
visual inspection 9
visualization techniques 9
water body 18
water-related changes 22
watershed modeling purposes 23
wavelength-based band spectrum 105
wetness
bands 91
direction 105
Window height 79
Wizard
Interface 35
Mode 53
Wizard interface 53
113
workspace
folder 101
workspace directory 6, 101
name 101
X
XML
file 32
tags 103
XML-based text file 103
Z
Zoom
Out 74, 75
Tool 70, 77
zoom
buttons 70
level 74, 77
114
Index
Index
115
116
Index
Index
117
118
Index