User manual MAVI -- Modular Algorithms for Volume Images

User manual MAVI -- Modular Algorithms for Volume Images
Quantitative analysis of microstructures
User manual
February 4, 2014
Fraunhofer-Institut für Techno- und Wirtschaftsmathematik ITWM, Abteilung
Bildverarbeitung, Fraunhofer-Platz 1, 67663 Kaiserslautern, Germany,
www.mavi-3d.de, [email protected]
Contents
Contents
1 MAVI’s purpose and scope
1
2 MAVI versions and licensing models
2
2.1 Demo version . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2
2.2 Full license . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2
3 Framework
2
4 Fundamental concepts
6
4.1 Images in MAVI . . . . . . . . . . . . . . . . . . . . . . . . . . .
6
4.1.1 Image attributes . . . . . . . . . . . . . . . . . . . . . . .
6
4.1.2 Image types . . . . . . . . . . . . . . . . . . . . . . . . .
7
4.2 MAVI’s coordinate system . . . . . . . . . . . . . . . . . . . . .
7
4.3 Discrete connectivity by adjacency systems . . . . . . . . . . . . .
7
4.4 Directions given by the cuboidal lattice . . . . . . . . . . . . . . .
8
4.5 Edge treatments . . . . . . . . . . . . . . . . . . . . . . . . . . .
9
5 First analysis
13
6 The typical MAVI workflow
13
7 Tutorials
15
7.1 Open cell nickel foam . . . . . . . . . . . . . . . . . . . . . . . .
16
7.1.1 Porosity and model based mean cell size . . . . . . . . . .
16
MAVI user manual
i
Contents
7.1.2 Cell size distribution via cell reconstruction . . . . . . . .
16
7.2 Closed cell zinc foam . . . . . . . . . . . . . . . . . . . . . . . .
20
7.3 Glass fiber reinforced composite . . . . . . . . . . . . . . . . . .
22
7.4 Sintered copper . . . . . . . . . . . . . . . . . . . . . . . . . . .
24
8 Reference manual
26
8.1 File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
8.1.1 Import . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
8.1.2 Export . . . . . . . . . . . . . . . . . . . . . . . . . . . .
32
8.1.3 Edit Header . . . . . . . . . . . . . . . . . . . . . . . . .
33
8.1.4 Close View . . . . . . . . . . . . . . . . . . . . . . . . .
33
8.1.5 Close Document . . . . . . . . . . . . . . . . . . . . . .
33
8.1.6 Close All Documents . . . . . . . . . . . . . . . . . . . .
33
8.1.7 Quit MAVI . . . . . . . . . . . . . . . . . . . . . . . . . .
33
8.2 Manipulation . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
8.2.1 Crop . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
8.2.2 AutoCrop . . . . . . . . . . . . . . . . . . . . . . . . . .
36
8.2.3 Extract by label . . . . . . . . . . . . . . . . . . . . . . .
36
8.2.4 Pad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
36
8.2.5 Shrink . . . . . . . . . . . . . . . . . . . . . . . . . . . .
37
8.2.6 Cast . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
37
8.2.7 Convert . . . . . . . . . . . . . . . . . . . . . . . . . . .
37
8.2.8 Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . .
37
8.2.9 Equalize . . . . . . . . . . . . . . . . . . . . . . . . . . .
38
MAVI user manual
ii
Contents
8.2.10 Prepare label image for visualization . . . . . . . . . . . .
39
8.2.11 Complement . . . . . . . . . . . . . . . . . . . . . . . .
39
8.2.12 Complex Conjugate . . . . . . . . . . . . . . . . . . . . .
40
8.2.13 Mask Image . . . . . . . . . . . . . . . . . . . . . . . . .
40
8.2.14 Mask with Cylinder . . . . . . . . . . . . . . . . . . . . .
41
8.2.15 Unary Operations . . . . . . . . . . . . . . . . . . . . . .
42
8.2.16 Binary Operations . . . . . . . . . . . . . . . . . . . . . .
43
8.3 Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . .
44
8.3.1 Binarization . . . . . . . . . . . . . . . . . . . . . . . . .
44
8.3.2 Hysteresis thresholding . . . . . . . . . . . . . . . . . . .
45
8.3.3 Niblack Segmentation . . . . . . . . . . . . . . . . . . . .
46
8.3.4 Sauvola Segmentation . . . . . . . . . . . . . . . . . . .
47
8.3.5 Labeling . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
8.3.6 Watershed Transformation . . . . . . . . . . . . . . . . .
47
8.3.7 Watershed Transformation, Preflooded . . . . . . . . . .
48
8.3.8 Cell Reconstruction . . . . . . . . . . . . . . . . . . . . .
54
8.3.9 Particle Separation . . . . . . . . . . . . . . . . . . . . .
55
8.4 Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . .
56
8.4.1 Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
56
8.4.2 Morphology . . . . . . . . . . . . . . . . . . . . . . . . .
56
8.4.3 Distance Transformation . . . . . . . . . . . . . . . . . .
62
8.4.4 Geodesic Transformation . . . . . . . . . . . . . . . . . .
65
8.4.5 Spectral Transformation . . . . . . . . . . . . . . . . . . .
72
MAVI user manual
iii
Contents
8.4.6 Cut Hills . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
8.4.7 Fill Holes . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
8.4.8 Extract Hills . . . . . . . . . . . . . . . . . . . . . . . . .
73
8.4.9 Extract Holes . . . . . . . . . . . . . . . . . . . . . . . .
73
8.4.10 Skeleton . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
8.4.11 Skeleton Analyzer . . . . . . . . . . . . . . . . . . . . . .
74
8.4.12 Object Filter . . . . . . . . . . . . . . . . . . . . . . . . .
74
8.4.13 Map Subfield Feature To Image . . . . . . . . . . . . . . .
75
8.4.14 Map Subfield Fiber Directions To Image . . . . . . . . . .
75
8.5 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
78
8.5.1 Image Statistics . . . . . . . . . . . . . . . . . . . . . . .
78
8.5.2 Rotation Mean . . . . . . . . . . . . . . . . . . . . . . .
78
8.5.3 Field Features . . . . . . . . . . . . . . . . . . . . . . . .
79
8.5.4 Subfield Features . . . . . . . . . . . . . . . . . . . . . .
82
8.5.5 Area Fraction Profile
. . . . . . . . . . . . . . . . . . . .
83
8.5.6 Grayvalue Profile . . . . . . . . . . . . . . . . . . . . . .
84
8.5.7 Granulometry . . . . . . . . . . . . . . . . . . . . . . . .
84
8.5.8 Object Features . . . . . . . . . . . . . . . . . . . . . . .
85
8.5.9 Open Foam Features . . . . . . . . . . . . . . . . . . . .
88
8.5.10 Geometric Tortuosity . . . . . . . . . . . . . . . . . . . .
89
8.5.11 (Sub-)Field Fiber Directions . . . . . . . . . . . . . . . . .
90
8.5.12 Spectral Methods . . . . . . . . . . . . . . . . . . . . . .
93
8.6 View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
95
MAVI user manual
iv
Contents
8.6.1 Slice View . . . . . . . . . . . . . . . . . . . . . . . . . .
95
8.6.2 Image Attributes View . . . . . . . . . . . . . . . . . . .
98
8.6.3 Volume Rendering View . . . . . . . . . . . . . . . . . .
98
8.6.4 Volume Rendering View - Selection . . . . . . . . . . . . 104
8.7 Help . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
8.7.1 System Info . . . . . . . . . . . . . . . . . . . . . . . . . 110
8.7.2 Help Assistant . . . . . . . . . . . . . . . . . . . . . . . . 110
8.7.3 About . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
9 Frequently Asked Questions
111
10 Optional modules
112
10.1 Particle Features . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
10.2 Point Field Statistics . . . . . . . . . . . . . . . . . . . . . . . . . 112
10.3 Mesh Export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
10.4 Mesh Vis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
10.5 Gray Value Mapping . . . . . . . . . . . . . . . . . . . . . . . . 116
References
118
Index
120
MAVI user manual
v
MAVI’s purpose and scope
1
MAVI’s purpose and scope
MAVI is a software for processing and analysis of volume images as produced
e.g. by micro computed tomography. It is a laboratory tool intended to
complement any device for 3D image acquisition.
MAVI is particularly suited for quantitative analysis of materials microstructures
like fiber reinforced composites, foams, non-woven, concrete, powders. Due to
its modular design and the universal basic geometric concepts used, MAVI is
capable to deal with three-dimensional images of other structures like bone or
snow too.
MAVI focuses on the characterization of the complex geometry of
microscopically heterogeneous structures. Volume, surface, integrals of
curvature and Euler number are determined for the whole structure or isolated
objects. Anisotropies and preferred directions are not only found but their
strength is measured, too. Special functions have been tailored for the analysis
of open foams or fiber reinforced composites. Metrology in the sense of
dimensioning of parts and comparison between actual and nominal dimensions
of a part is not in MAVI’s scope.
MAVI’s analysis core is complemented by various image processing and
segmentation techniques. 2D slices provide fast visual control of processing
steps while the volume rendering yields a spatial visualization.
This handbook intends to
• give the non-expert user a starting point to get into MAVI
• give the experienced user detailed information on the algorithms used
• provide a reference for finding the appropriate processing chain for a
particular problem.
MAVI user manual
1
MAVI versions and licensing
models
2
MAVI versions and licensing models
2.1
Demo version
A demo version of MAVI is available via www.mavi-3d.de (center column, first
item in download list).
The demo version is fully functional except all functions for saving or exporting
image or analysis data being disabled.
The image size you can work is restricted to 1283 pixels,however. The
Import:Image Raw Data dialog offers the possibility to Crop such a sniplet out
of your 3D image data.
2.2
Full license
MAVI’s current licensing model is a single user floating license controled by an
USB dongle. To run a fully licensed MAVI, you need this special dongle, the
matching license file, and the MAVI installer (Windows) or rpm (Linux). Please
contact [email protected] to get information on how to obtain them.
If the dongle or a valid license file is missing, MAVI issues a warning and
proceeds in demo mode.
3
Framework
MAVI has a multi document interface, which allows for processing of multiple
documents at the same time. The application main window provides menus, a
toolbar, several docking windows and a statusbar surrounding a large central
workspace, see Figure 1.
The main menu provides access to the algorithms. See Sections 8.1-8.7 for
detailed descriptions of the groups:
MAVI user manual
2
Framework
Figure 1: MAVI’s main window.
File
Manipulation
Segmentation
Transformation
Analysis
View
Help.
A menu entry is active (i.e. can be chosen by clicking on it) only if the
corresponding function can be applied to the selected image (that is, if the
image is of a valid gray value type and size).
•
•
•
•
•
•
•
The toolbar
offers
control functionality for the current view window. For more details, please refer
to the Slice View Section 8.6.1.
The central workspace contains all view windows available in the application.
Each view is associated with a specific document and displays its content. The
MAVI user manual
3
Framework
foremost view of the workspace is assigned to the active document. All
operations will be applied to this document. Activate a window by clicking on it
or its symbol in the view navigator window. It is possible to have different kinds
of views for one document at the same time. By default, each view is
maximized. That means the view fills the complete workspace when it appears
for the first time.
Navigation in the view navigator window by cursor is not possible. Under
Windows operating systems this can however easily be done using Ctrl and Tab
keys.
The titlebar
of a view window consists of the view icon on the left hand side, which provides
a window menu when clicked, the name of the view window, and three control
buttons on the right hand side. The window menu allows you to close, resize
and move the window. The
and
buttons minimize and miximize the
window, respectively, and the
button closes it. If the last view window of a
document is closed, the document is closed too. When a window is maximized,
the view icons and the control buttons are displayed on the left and right hand
side of the menu bar
In this case, the
button turns into a
size of the window when clicked.
.
button which restores the original
At startup the MAVI framework has two windows docked on the left border of
the workspace, a view navigator (Figure 2(b)) on the top and a view control
window below. The windows can be moved around by using the mouse on the
docking edge (Figure 2(a)). If the window is not docked to a side of the
workspace, then it becomes a floating standalone window. By clicking on a
docking window or on the menu bar with the right mouse button you obtain a
control list of all available docking windows which allows to activate and
deactivate them.
The view navigator window shows a list of all available views. A list entry
consists of the view icon, the view label and an additional description provided
by the view. The active entry is marked by a different background color. When
an entry is selected, the assigned view becomes active and is moved to the
foreground of the workspace.
The view control window provides several functions and information for the
current view. Its content is different for different kinds of views. See
Sections 8.6.1- 8.6.3 for details on the particular view types and Figure 45 for
the view control window of the Slice View.
MAVI user manual
4
Framework
(a) Docking Edge
(b) View Control
Figure 2: MAVI’s framework: Docking Edge and View Control Window.
MAVI user manual
5
Fundamental concepts
4
Fundamental concepts
To take advantage of MAVI’s full potential, a basic understanding of the
following concepts is needed. However, the material presented here is certainly
not exhaustive. Thus references to the relevant literature are provided for the
user interested in more background information.
4.1
Images in MAVI
MAVI is built on the idea of pixels (or voxels) being the vertices of a grid and
holding information on the image content as gray values. More precisely, let
L = (s1 , s2 , s3 )Z3 be a three-dimensional cuboidal lattice with lattice spacings
s1 , s2 , s3 > 0. For data from micro-computed tomography, it holds usually that
s1 = s2 = s3 and with a slight abuse of notation one speaks of an “isotropic”
lattice. Analogously, lattices with different spacings are refered to as
“anisotropic” lattice.
MAVI’s core analysis functions as well as essential transforms like the Euclidean
distance tranform, work for image data with pairwise different lattice spacings
e. g. from so-called FIB tomography, too. However, filters as well as
morphological transforms are pixel based and thus do not correct for
“anisotropic” lattice spacings.
4.1.1
Image attributes
An image is not just described by its pixel’s gray values. MAVI stores the
following image attributes in the image header:
Creator A string containing information about the image creator.
Description A string containing information about the content of the image.
History A string summarizing the transforms applied to the image.
Offset A triple of integer values allowing to relate to some global coordinate
system.
Spacing A triple of floating point values describing the physical size of pixel
spacings or edges in [m].
The image header entries will either be created when loading an image from a
file 8.1 containing the respective information or can be edited by the user via
the Edit Header function. Before applying analysis functions, it should always
MAVI user manual
6
Image types
be ensured that the image spacings are set correctly. If no spacing is set, the
base unit for all measurements is [Pixel].
4.1.2
Image types
So far, MAVI deals with scalar images only. The only exception from this rule are
COMPLEX images as arising from the Fast Fourier Transformation.
image type
MONO
GRAY8
GRAY16
GRAY32
GRAYF
COMPLEX
description
just 2 components –
foreground and background
8 bit integer gray values
16 bit integer gray values
32 bit integer gray values
floating point gray values
(single IEEE754 precision)
complex floating point gray values
(single IEEE754 precision)
range
0 (background), 1 (foreground)
[0, 255]
[0, 65.535]
[0, 4.294.967.295]
approx. 1.17e−38 , 3.40e+38
2
approx. 1.17e−38 , 3.40e+38
for both real and imaginary components
Table 1: MAVI’s image types
4.2
MAVI’s coordinate system
MAVI has a right-handed coordinate system and the origin (0, 0, 0) is
interpreted to be the lower left front corner of the cuboid representing the
observation window, see also Slice View.
When importing image data MAVI assumes the data to be X local if not
explictely stated otherwise by the original image format. That is, the data is read
with the X coordinate running the fastest and the Z coordinate the slowest
resulting in e. g. pixel (1, 0, 0) being read before pixel (0, 0, 1).
4.3
Discrete connectivity by adjacency systems
Usually, discrete neighborhood graphs are used to define how the pixels of an
image are connected. However, in 3D, neighborhood graphs do not yield a
unique description anymore. Therefore MAVI uses adjaceny systems consisting
of vertices, edges, faces, and cells instead. For 3D images, the adjacency
systems described in [12] are used – 6, 14.1, 14.2, and 26 adjacency system, see
Figure 3 for a sketch. For 2D images, MAVI uses the 4, 6.1, 6.2, and 8
adjacency systems.
MAVI user manual
7
Directions given by the cuboidal
lattice
(a) 6 adjacency system
(b) 14.1 adjacency system
(c) 14.2 adjacency system
(d) 26 adjacency system
Figure 3: 3D adjacency systems, direct neighbors (green) of the central pixel
(blue).
There is no overall best adjacency system, see [10] for a rigourous proof. It is
important to be consistent, in particular to segment using the same adjacency
system as used later on e. g. for a connectivity analysis. When aiming at
separating image components, the 6 or 4 adjacency systems are a better choice
than the 26 or 8 adjacency systems, respectively. It’s exactly the other way round
when the goal is to preserve thin connections.
4.4
Directions given by the cuboidal lattice
The unit cell of the cuboidal lattice determines a number of discrete directions.
MAVI uses two sets of 13 (3D) or 8 (2D) directions for morphological transforms
8.4.2 and some global analyses provided by the Field Features and Subfield
Features. One set of 13 directions in 3D is given by lines connecting vertices of
the lattice unit cell – 3 coordinate directions, 6 face diagonal directions, 4 space
diagonal directions as specified in the following Table 2. For an illustration see
Figure. The spacing vector (s1 , s2 , s3 ) is the one introduced in 4.1 above.
The second set of 13 directions in 3D, used by Field Features and Subfield
Features are the normal directions to planes defined by vertices of the unit cell.
In the case of an isotropic lattice, the two sets of directions coincide, just the
MAVI user manual
8
Edge treatments
Direction
00
01
02
03
04
05
06
07
08
09
10
11
12
vector
(1, 0, 0)
(0, 1, 0)
(0, 0,p
1)
(s1 , s2 , 0)/ ps21 + s22
(−s1 , s2 , 0)/p s21 + s22
(s1 , 0, s3 )/ ps21 + s23
(−s1 , 0, s3 )/p s21 + s23
(0, s2 , s3 )/ ps21 + s23
(0, −s2 , s3p
)/ s21 + s23
(s1 , s2 , s3 )/ ps21 + s22 + s23
(−s1 , s2 , s3 )/ps21 + s22 + s23
(s1 , −s2 , s3 )/ ps21 + s22 + s23
(−s1 , −s2 , s3 )/ s21 + s22 + s23
description
parallel to the x-axis
parallel to the y-axis
parallel to the z-axis
face diagonal
face diagonal
face diagonal
face diagonal
face diagonal
face diagonal
space diagonal
space diagonal
space diagonal
space diagonal
Table 2: Discrete directions given by the unit cell of the cuboidal lattice.
order differs. However, for anisotropic lattices this does not hold and therefore
the normal directions are listed in Table 3 below.
For 2D images, the 8 directions – 4 coordinate directions and 4 diagonal
directions – are given by their angle to the x-axis in radians.
4.5
Edge treatments
Image transforms like filters or morphological transforms have to apply some
kind of edge treatment for determining the new values of the pixels at or close
to the image borders.
The following table lists the different edge treatments offered by MAVI’s
algorithms. The initials are those appearing in the image history, see 4.1.1. The
choice of available edge treatments depends on the respective algorithm.
Remark 1 The edge treatments given here refer to image transforms only. See
Section 8.5 for the edge correction options of the analysis algorithms.
MAVI user manual
9
Edge treatments
Direction
00
01
02
03
04
05
06
07
08
09
10
11
12
vector
(0, 0, 1)
(0, 1, 0)
(1, 0, p
0)
(0, −s3 , s2 )/p s22 + s23
(0, s3 , s2 )/ ps22 + s23
(−s3 , 0, s1 )/p s21 + s23
(s3 , 0, s1 )/ps21 + s23
(s2 , s1 , 0)/ ps21 + s22
2
2
(−s2 , sp
1 , 0)/ s1 + s2
(−s2 s3 , −s1 s3 , s1 s2 )/p (s1 s2 )2 + (s1 s3 )2 + (s2 s3 )2
(s2 s3 , −s1 s3 , s1 s2 )/p (s1 s2 )2 + (s1 s3 )2 + (s2 s3 )2
(s2 s3 , s1 s3 , s1 s2 )/ p(s1 s2 )2 + (s1 s3 )2 + (s2 s3 )2
(−s2 s3 , s1 s3 , s1 s2 )/ (s1 s2 )2 + (s1 s3 )2 + (s2 s3 )2
description
normal to xy-plane
normal to xz-plane
normal to yz-plane
normal to plane cutting 4 vertices
normal to plane cutting 4 vertices
normal to plane cutting 4 vertices
normal to plane cutting 4 vertices
normal to plane cutting 4 vertices
normal to plane cutting 4 vertices
normal to plane cutting 3 vertices
normal to plane cutting 3 vertices
normal to plane cutting 3 vertices
normal to plane cutting 3 vertices
Table 3: Discrete normal directions given by the unit cell of the cuboidal lattice.
Edge Treatment
No Edge Treatment
Embedded
Periodic
Reflective
Reflective Double Edge
Description
No edge treatment employed
The image is embedded into background; the background value
is set by the algorithm.
The image data is extended periodically.
The image data is extended by reflection on the border pixels.
The image data is extended by reflection outside the image borders, i.e. border pixels are repeated.
Abbrev.
NET
EMB
PER
REFL
RDE
Table 4: Edge treament options for image transforms.
MAVI user manual
10
Edge treatments
(a) 00
(b) 01
(c) 02
(d) 03
(e) 04
(f) 05
(g) 06
(h) 07
(i) 08
(j) 09
(k) 10
(l) 11
(m) 12
(n) Coordinate system
Figure 4: Discrete directions given by the unit cell of the cuboidal lattice.
MAVI user manual
11
Edge treatments
(a) 00
(b) 01
(c) 02
(d) 03
(e) 04
(f) 05
(g) 06
(h) 07
(i) 08
(j) 09
(k) 10
(l) 11
(m) 12
(n) Coordinate system
Figure 5: Discrete normal directions given by the unit cell of the cuboidal lattice.
MAVI user manual
12
First analysis
5
First analysis
In order to facilitate the very first steps with MAVI, this section describes step by
step how to solve a very simple yet typical analysis task: Determining the
specific surface area of a closed cell metal foam.
(1) Start MAVI. The window is empty except for the title bar, the main menu,
and the status bar on the bottom.
(2) Open the demo image AlFoam128.iass by choosing Open Image from
the File menu. MAVI opens a dialog for specifying path and file name.
(3) The Slice View now shows the first slice in each coordinate direction.
Clicking into one of the three views and subsequently moving of the slider
in the toolbar above enables navigation through the slices.
(4) Choosing Binarization from the Segmentation menu opens a dialog
box with preview for choosing a global gray value threshold. Ticking
“Whole image histogram” activates the button “Otsu’s threshold”.
Clicking that one in turn yields a lower threshold of 34. OK closes the
dialog and creates a new image appearing in the document list as
AlFoam128.iass 2.
(5) Now the Field Features from the Analysis menu can be applied and the
view switches to an attribute view.
(6) Extending the attributes tree by clicking the Field Features and the Surface
entries now shows the specific surface area (“surface density”) measured
in [1/m].
6
The typical MAVI workflow
The typical workflow contains a lot more steps and is sketched in the following:
(1) Import image data using File:Import. Either the file type needed is listed
there or the generic importing functions Image Raw Data or Image
Stack have to be used.
(2) Check whether the pixel spacings are set correctly by View:Image
Attributes View and extending the spacings subtree there. Correct the
spacings if needed using File:Edit Header.
MAVI user manual
13
The typical MAVI workflow
(3)
Navigate through the data sets using the slice views or the Slice Show to
get an impression of the structure. Use different view filters if needed.
(4) Reduce the observation window to the region containing structural
information by Manipulation:Crop
(5) Remove noise using e. g. the Median Filter from Transformation:Filter
(6) Segment the component of interest by one of the binarization methods
from the Segmentation menu.
(7) Obtain global geometric characteristics of the foreground component by
computing the Field Features.
(8) Export the analysis results using File:Export:Field Features as CSV
(9) Import the csv-file into your preferred statistics package like Excel or R.
Ensure that “;” is used as separator and “.” as decimal point.
Further processing, segmentation, and analysis steps will follow depending on
the application. The tutorials presented in the following Section 7 cover the
most common cases.
MAVI user manual
14
Tutorials
7
Tutorials
This section offers several step-by-step tutorials for frequently occuring analysis
tasks.
MAVI user manual
15
Open cell nickel foam
7.1
Open cell nickel foam
This tutorial consists of two parts. The first one demonstrates how to measure
mean features of an open foam like porosity, specific surface area, or mean cell
size based on the original gray scale image obtained by computed tomography
(CT). In particular, the tutorial shows all processing steps neccessary before the
actual analysis.
The second part of the tutorial shows how to measure the empirical size
distribution of the foam cells by image analytical cell reconstruction.
7.1.1
Porosity and model based mean cell size
Figure 6 shows slices from the reconstructed CT image of an open cell nickel
foam:
(1) Clearly, the structure does not fill the image, but is surrounded by a lot of
empty space along the image borders. These empty areas affect the Field
Features’ result and therefore have to be removed using
Manipulation:Crop.
(2) Segmentation:Binarization with Lower Threshold 70 and Upper
Threshold 256 segments the solid component.
(3) Some small foreground components not connected to the solid structure
are left. Segmentation:Labeling identifies them. Manipulation:Extract
by label with label 1 yields a binary image without these crumbs.
(4) Now the Analysis:Field Features yield porosity, specific surface area, and
various other global characteristics.
(5) The mean cell size is deduced from the Euler number density and a (mild)
model assumption [18]. Thus the pores inside the struts have to be closed.
To this end Transformation:Morphology:Closure with structuring
element Cube and size 3 fills the small holes inside the edges of the foam.
(6) Another Labeling, this time applied to the background (“apply to
foreground” NOT ticked) and Extract by label with label 1
Now Field Features and Open Foam Features from the Analysis menu yield
porosity and specific surface area and mean cell size, respectively.
7.1.2
Cell size distribution via cell reconstruction
In order to deduce empirical distributions of cell characteristics, the pore space
has to be divided. To this end, apply Segmentation:Cell Reconstruction by
Complex Morphology to the result of Section 7.1.1 shown in Figure 6(h). Use
Dynamic 19 to obtain the cell system as shown in Figure 7
MAVI user manual
16
Cell size distribution via cell
reconstruction
Now the Object Features compute a variety of geometric characteristics of the
reconstructed cells. To export them, choose Export:Object Features as CSV
form the File menu. Import the csv-file into your preferred statistics package
like Excel or R. Ensure that “;” is used as separator and “.” as decimal point.
Make sure to correct for edge effects when estimating means, standard
deviations, or other moments of the respective distributions.
MAVI user manual
17
Cell size distribution via cell
reconstruction
(a) original
(b) cropped
(c) binarized
(d) labeled
(e) crumbs removed
(f) closed
(g) background labeled
(h) inner pores removed
Figure 6: Slices through the open nickel foam image illustrating the tutorial steps.
MAVI user manual
18
Cell size distribution via cell
reconstruction
(a) cells reconstructed
Figure 7: Slice through the open nickel foam image after cell reconstruction.
MAVI user manual
19
Closed cell zinc foam
7.2
Closed cell zinc foam
This tutorial shows how to reconstruct foam cells if the cell size varies strongly.
A closed cell zinc foam sample is used. The 3D image data set was obtained by
computed tomography at the BAM line of BESSY. The pixel spacing is 5.5µm.
Both sample and image data are courtesy of the Helmholtz-Zentrum Berlin.
Figure 8 uses x-y-slice 126 to illustrate the tutorial steps:
(1) Load kombin172-171.orig.iass.gz (a)
(2) Reduce noise by applying Transformation:Filter:Median filter (b)
(3) Use global thresholding (Segmentation:Binarization) with lower
threshold 98 to binarize (c)
(4) Apply the Transformation:Distance Transformation:Euclidean
Distance Transformation on the background (no edge treatment, exact
values)
(5) Transform the resulting float values into integers by
Manipulation:Spread with Output type 8bit integers, Grayvalue
characteristic linear, and Spread from grayvalue range of interest? not
ticked (d)
(6) Smooth the image by
Transformation:Geodesic Transformation:Adaptive H-Extrema
choosing H-Maxima, 26 Neighborhood, Dynamic 3 at 0 and 11 at
maximum
(7) Add the value 1 to all pixels by Manipulation:Unary Operations
(8) Use Manipulation:Complement to invert the binary image obtained in
step (3) (e)
(9) Use this binary image to mask the smoothed lifted distance image:
Choose Manipulation:Mask Image and the result from (7) as image that
should be masked. These three step ensure that the foam structure has
everywhere in the image strictly lower gray values than the pore space.
(10) Invert by Manipulation:Complement to turn the cell centers into local
minima (f)
(11) Apply Segmentation:Watershed Transformation with 26
neighborhood (g)
(12) Return to the mask image (8), mask the watershed result with this (h)
MAVI user manual
20
Closed cell zinc foam
(a) original
(b) median filtered
(c) binarized
(d) distance transform
(e) mask
(f) inverted smoothed distance transform
(g) watershed transform
(h) masked watershed transform
Figure 8: Slices through the closed zinc foam image illustrating the tutorial steps.
MAVI user manual
21
Glass fiber reinforced composite
7.3
Glass fiber reinforced composite
This tutorial shows how to analyse the fiber orientations in a fiber reinforced
composite. The sample provided by Institut für Verbundwerkstoffe (IVW)
Kaiserslautern features long fibers and a fiber volume fraction above 50%. The
3D image data set with a pixel spacing of 3.0µm was obtained by computed
tomography at the IVW as well.
Figure 9 illustrates the tutorial steps using x-y-slice 250:
(1) Load GF60-crop512.iass.gz (a)
(2) Explore the gray value distribution by the Analysis:Grayvalue Profile.
The global decline of the mean gray value in y-direction will cause global
thresholding methods to fail. Therefore a so-called shading correction will
be performed in the following steps.
(3) Create a strongly smoothed version of the input image by
Transformation:Morphology:Opening, choosing “Cube” as
structuring element (3D Struct. Element) with size 60 (b)
(4) Subtract the smoothed image from the original one by returning to the
original image and applying
Manipulation:Binary Operations:Subtract. Choose
GF60-crop512.iass.gz 2 as second image for Subtracting (the
subtrahend). The resulting image is much darker as gray value dynamic
gets lost, however, the mean gray value in y-direction changes only
mildely now. (Check using the Grayvalue Profile again.) (c)
(5) Now the (Sub-)Field Fiber Directions from the Analysis menu can be
applied.
(a) The glass fibers are known to have a cross section diameter of
approximately 10µm. Thus the fiber radius is 1-2 pixels. Select
6e-06 (6 · 10−6 m = 6µm) corresponding to 2 pixels.
(b) Tick “Select subfield size”. Enter 135e-06 (135 · 10−6 m = 135µm)
corresponding to 45 pixels for x, accept this value for y and z, too.
(c) Tick “Advanced options”. Change the “Otsu factor” from its
default value of 1.25 to 1.0. (Due to the shading correction.) Push
“OK”. This step takes a couple of minutes.
(6) Use Transformation:Map Subfield Fiber Directions To Image to
visualize the rheological effects. Select the Orientation tensor/yy. This is
the diagonal entry of the orientation tensor in y-direction (a22 from (15)).
Tick “Crop result”. Push OK. (d)
(7) A color view filter would increase the visual effect. However, to apply a
View Filter, the resulting GRAYF image has to be Spread (Manipulation)
to 8bit integer gray values.
(8) Now apply the Power Spectrum (blue -> red) (e) color table.
(9) Return to the attributes view and File:Export:(Sub-)Field Fiber
Directions as CSV.
MAVI user manual
22
Glass fiber reinforced composite
(10)
Import the csv-file into your preferred statistics package like Excel or R.
Ensure that “;” is used as separator and “.” as decimal point. Plot
“Orientation tensor/xx”, “Orientation tensor/yy”, “Orientation tensor/zz”
over “slice position”. (f)
(a) original
(b) opened
(c) shading corrected
(d) orientation tensor diagonal entry in ydirection
(e) orientation tensor yy, colored
(f) orientation tensor profile plot
Figure 9: Slices through the GFRP image illustrating the tutorial steps.
MAVI user manual
23
Sintered copper
7.4
Sintered copper
This tutorial shows how to separate interconnected particles in order to analyze
them indivdually. The sample provided by Institute of Materials Science at
University of Dresden was obtained by computed tomography at the BAM with
a pixel spacing of 14µm, see [9] for more background information.
Figure 10 illustrates the tutorial steps using x-y-slice 125:
(1) Load s060.iass.gz (a)
(2) Before starting to work on the sintered copper, the balls have to be
separated from the cylindrical crucible holding them. This can be easily
achieved choosing the lower gray value threshold for
Segmentation:Binarization appropriatly as the aluminium crucible is
darker (has lower gray values) than the copper balls. Figure (2)(a) shows
how this is appears in the Segmentation:Binarization preview window.
(a) Binarization dialog
(3)
(4)
(5)
(b) Crop dialog
Since there is much space without any information in the resulting image,
crop by Manipulation:Crop with parameters choosen as in the dialog
displayed in Figure (2)(b). The resulting image should now have a size of
333x333x230 pixels.
An attempt to binarize by Segmentation:Binarization with lower
threshold value 150 shows a quite rough structure, see Figure (2)(b).
Therefore, choose the previous image (after cropping) and apply
Transformation:Filter:Median Filter on it. Choose "Double Reflective
Edge" as Edge Treatment and a Filter Size of 3.
Now binarize the result by Segmentation:Binarization with lower
threshold value 150 Figure 10(c). The resulting surface is much smoother
MAVI user manual
24
Sintered copper
(6)
(7)
(8)
(9)
and the image is suitable for particle separation.
MAVI provides automatic particle separation in the Segmentation menu.
Apply Particle Separation by Preflooded Watershed with an area
threshold of 20 (Figure 10(d))
Compute the Analysis:Object Features for the result. Do not exclude
the background. If the "Object Statistics" are computed does not matter
if you postprocess the analysis data externally.
Export the Object Features using Object Features as CSV from the
Export submenu of the File menu.
Import the csv-file into your preferred statistics package like Excel or R.
Ensure that “;” is used as separator and “.” as decimal point. Plot the
frequency of some statistics, e.g. “Diameter” (Figure 10(e)). Note that in
this particular case, no edge correction is needed as all balls are
completely contained in the image.
(a) original
(b) cropped
(d) particle separation
(e) diameters frequency plot
(c) binarized
Figure 10: Slices through the sintered copper image illustrating the tutorial steps.
MAVI user manual
25
Reference manual
8
Reference manual
MAVI user manual
26
File
8.1
File
The File menu provides load and save function for MAVI’s native 3D image data
format (.iass or .iass.gz) and various 2D image formats as well as generic import
and export nodes for 3D image data either given as a block of raw data or as a
stack of 2D images. The Export submenu subsumes various functions for
saving Analysis results.
When opening or importing a file, MAVI chooses the last path used for opening
or importing as default path. The last path used in a specific MAVI session will
be stored and used in the next session.
If you work with the demo-version of MAVI, MAVI restricts the size of the
images you are able to work with to 1283 pixels. It is nevertheless possible to
open parts of larger images by cropping them to at most 1283 pixels. All save
and export functions are however disabled in the demo-version.
Remark 2 Image meta data as available via Edit header is completely loaded
and saved when using MAVI’s .iass or .iass.gz formats. When importing or
exporting other image data formats, make sure to provide and store essential
meta data, respectively. Check the description of the functions for details.
Open Image This function opens IASS and IASS.GZ files and 3D TIFF files as
3D images in MAVI. The supported 2D file types are BMP, TIFF, PNG, PBM, PGM,
PXM, JPG, and JP2 for 2d data.
3D TIFF does not contain information about the pixel spacing in z-direction.
Therefore please use File:Edit header to insert the correct pixel spacings.
Otherwise, all Analysis results for this image will have [pixel] as base unit.
Save Image This function saves the current image as IASS, IASS.GZ, or 3D
TIFF in case it is a 3D image, and as PGM, BMP, or TIF in case it is a 2D image.
The path suggested for saving is the one used for the last opening of a file.
Image meta data as available via Edit header is saved in IASS(.GZ), only.
To save a slice of a 3D image as a 2D image, there are two options:
(1) Save the selected slice using the “Save current slice view” option from the
“Settings” toolbox of the view control window of the Slice View. This
saves the view to the slice, not the slice data itself.
(2) Crop the slice of thickness one pixel from the 3D image. Use Save Image
and choose explicitly “2d Image” in the file type entry of the dialog. This
option works for slices in the XY plane only.
MAVI user manual
27
Import
8.1.1
Import
AVS Data The AVS field data is assumed to be given by two files: a data file
of type DAT or REK and an ASCII description file of type FLD which specifies the
name of the file containing the actual image data and defines the structure of
the AVS field.
Input for Import:AVS Data is just the FLD file.
This is one of various possible representations of AVS field data. For more
information see www.avs.com.
Image Raw Data MAVI is capable of importing a large number of additional
formats, given they are saved to disk uncompressed. This is done generically,
meaning that virtually any file saved in such a format can be opened in MAVI.
The user has to specify a few parameters to help MAVI understand the format
structure. Such an initially unsupported format type is referred to as Raw Data
files. Common file suffixes for raw data files are RAW, BIN, VOL, REC, and REK.
Select the desired file and accept. A dialog will appear giving options to
fine-tune the import process. MAVI guesses the parameters. However, as
shown in Figure 11(a), this first guess is very likely to be wrong. In particular,
MAVI’s guess for the image size is based on the assumption of a cubic
observation window. Now the user should provide step by step more
information, until the preview window shows a reasonable content. Parameters
include the image dimensions, image color type, image orientation, data format
and the image header size.
To change a parameter, activate the checkbox next to it. Once a parameter is
active, MAVI no longer tries to guess its value. If values contradicting previous
input or the data are specified, MAVI marks the respective input parameters
with a red “!”. See Figure 11(b) for an example. These values have to be
corrected before importing the image. Once the preview shows the desired
image Figure 11(d), click Open. The parameters and their meaning are
summarized in Table 5. Please make sure to fill in the pixel spacings correctly as
all Analysis results will be based on the spacings. Please note that the base unit
in MAVI is always [m]. The spinbox allowing to change the unit is just provided
for convenient input of very small nominal resolutions.
After importing the raw image data, pixel spacings can still be corrected using
Edit header in the File menu.
MAVI user manual
28
Import
(a) Default
(b) Size impossible
(c) Size correct
(d) Color type correct
Figure 11: Import:Image Raw Data: dialog.
Dicom Data In the open file dialog, see Figure 12(a), choose either a DICOM
header file of type DIC or DCM or a DICOM dump file of type TXT and click
Open. A list of available 2d image files will be shown, see Figure 12(b). The
number of files selected should coincide with the number given in the DICOM
header file. However, in most cases MAVI is able to open and display the
DICOM data correctly even if that is not the case.
In the preview dialog, check if all data are correct and click Open to accept, see
Figure 12(c). The tags found in the DICOM header are shown in the table at the
top. For details on a specific entry, simply click on it. The information will then
be shown in the lower right window. The window in the lower left corner
shows a preview of a slice of the actual 3D image. To check the number of 2D
files belonging to the DICOM dataset is given in the Section Count entry of the
tag table.
MAVI user manual
29
Import
Tab
Essential info
Parameter
Dimensions X/Y/Z
Order
Color Type
Treat Data As Signed Data
Swap Data Endianness
Treat Data As ASCII
Header Length
Orientation
Crop
Data Length
Original Locality
Offset
Size
Header Fields
Spacing
Description
Description
image size in pixels
specifies whether dimensions are given depth-first (DHW) or
width-first (WHD)
image type, supported are MONO, GRAY8, GRAY16,
GRAY32, and FLOAT
check if image data contains negative gray values
check if byte-order of the image data is big-endian, e. g.
coming from a SUN or Power-PC system, button remains inactive, if MAVI fails to interpret the data as ASCII
interpret input stream as ASCII
length of the header in bytes, usually correctly filled in by
MAVI once the Dimensions have been entered
length of the data in bytes, usually correctly filled in by MAVI
swap coordinate axes if needed
For large data sets, it might be sensible to crop before actually
importing.
defines the new lower left front corner of the image w. r. t.
the original data
defines the size of the part of the image to be imported
The memory requirement is shown in the lower part of the
dialog window.
set the pixel spacings
allows to enter image meta information like Creator, detailed
Description, and processing History
Table 5: Parameters of the Import:Image Raw Data: dialog.
After importing, the spacings of the resulting 3D image should be checked.
Otherwise, all Analysis results may be incorrect.
Image Stack This generic importing function loads 3D image data provided as
stack of 2D slices. In the section dialog, use the Shift key to select the desired
2D image files. Alternatively, try Ctrl A to select all and remove the unwanted
ones using Ctrl. Press OK to load the files in ascending alphanumerical order.
Supported file types are BMP, (uncompressed) TIFF, PGM, PBM, PXM, JPG, and
JP2.
All of the selected files must be two dimensional, and in one of the supported
formats BMP, (uncompressed) TIFF, PGM, PBM, PXM, JPG, and JP2. The images
must all be of equal dimension. Otherwise, the import will aborted.
MAVI user manual
30
Import
(a) Select header
(b) Select slices
(c) Preview dialog
Figure 12: Importing DICOM data: selecting a DICOM header file, selecting slice
images, and chekcing the import via the preview.
After importing the image stack, please use Edit header in the File menu to
insert the correct pixel spacings. Otherwise, all Analysis results will be in [pixel].
Fraunhofer GeoDict Volume (GDT) Loads the specified file and sets the
pixel spacings correctly.
Fraunhofer Raw Volume Data (REK) Loads the specified file and sets the
pixel spacings correctly. Note that the Fraunhofer Raw Volume Data is based on
a coordinate system different from MAVI’s. This results in the data being flipped
along the Z axis.
3D Electron Density File (MRC) The user has to specify the file and to input
the pixel spacings.
Just a subset of the various possible MRC file format versions is supported. The
pixel grey value type has to be
• little endian and
• 8 bit unsigned or
• 16 bit unsigned and signed or
• float.
In all other cases MAVI will not recognize your MRC file. Then please use
Import:Image Raw Data.
MAVI user manual
31
Export
8.1.2
Export
This submenu offers functions for exporting/saving image data in a choice of
image formats or as CSV (comma separated value) and analysis results as CSV
files. The CSV files use the semicolon as separator, the decimal separator is the
decimal point. Additionally, comment lines marked by a # can appear at the
beginning of the CSV file.
The following export functions for image data are available:
• Image as CSV (Text)
• Image as Fraunhofer GeoDict Volume (GDT)
• Image as Fraunhofer Raw Volume (REK)
• Image as AVS Data
• Image as Raw Data
Note that all four image formats store the pixel spacings but no other image
meta data. The CSV files contain nothing but the gray values. For the two
generic exporting functions Export:Image as CSV and Export:Image as Raw
Data, the gray values of the image’s pixels are written to the file with the X
coordinate running the fastest and the Z coordinate the slowest. That is, pixel
(1, 0, 0) is written before pixel (0, 0, 1). For MAVI’s coordinate system see
Section 4.
The following export functions for analysis results are available:
• Image Atrributes as CSV (Text)
• Field Features as CSV (Text)
• Area Fraction Profile as CSV (Text)
• Grayvalue Profile as CSV (Text)
• Granulometric Data as CSV (Text)
• Object Features as CSV (Text)
• Open Foam Features as CSV (Text)
• Geometric Tortuosity as CSV (Text)
• (Sub-)Field Fiber Directions as CSV (Text)
The export function for the Field Features offers the choice between global
and local features, that is between Field Features and Subfield Features, see
the dialog in Figure 13 below.
Figure 13: Export dialog for Field Features and Subfield Features as CSV.
MAVI user manual
32
Edit Header
Finally, there are export functions for rotation means of analysis results:
• Rotation Mean of Image Data as CSV (Text)
• Rotation Mean of Covariance as CSV (Text)
• Rotation Mean of Bartlett Spectrum as CSV (Text)
8.1.3
Edit Header
This function allows to set or change image meta information as shown by
Image attributes, see Section 4.1.1 for a list of fields. When saving the image
as IASS or IASS.GZ, this meta data is stored, too.
The pixel spacings in the header dialog can be given in micrometers (default),
meters, millimeters or nanometers. In the file header, however, it will always
appear in meters. Make sure the pixel spacings are set correctly as all
measurements by functions like Field Features or Object Features from the
Analysis menu use these spacings.
8.1.4
Close View
Closes the current view. If it is the only view on the respective document, then
the documentitself is closed.
8.1.5
Close Document
Closes the current document together with all view windows associated with
the document.
8.1.6
Close All Documents
Closes all documents listed in the document view together with all view
windows associated with these documents.
8.1.7
Quit MAVI
Closes MAVI together with all open documents.
MAVI user manual
33
Manipulation
8.2
Manipulation
The manipulation menu contains tools to manipulate the observation window
of an image and switching the coordinate axes. Moreover, a node for masking
an image with a MONO image is provided.
This menu provides basic unary and binary operations to be applied pixelwise
e.g. for combining or comparing images. The complement turning local minima
into maxima and vice versa is an indispensable tool when feeding distance data
to the Watershed Transformation.
8.2.1
Crop
Cropping an image means to decrease its size by cutting off slices. In the MAVI
cropping dialog, three slice views are visible, orthogonal to each of the three
coordinate axes. The part of the image to be cut out can be specified either
using the sliders beside and below these views or by changing the image size in
each of the coordinate directions given in the number boxes next to the sliders.
The intervals defining the cropping limits are half open, including the lower but
excluding the upper boundary. That is, to select exactly e.g. slice number 128
you need to choose lower boundary 128 and upper boundary 129.
The remaining part of the image is marked in red. The rectangle is filled iff the
current slice intersects the new, smaller image, too.
If a particular image size after cropping is desired, then
(1) first type in the upper bounds,
(2) activate the last one of them by clicking into one of the the slice views and
(3) position the marked cuboid by moving the centers of the sliders.
The buttons
and
allow to zoom in to get a close-up view of the image
slice or to zoom out to see more of the slice at a reduced size, respectively.
The crop control panel in the bottom right part of the crop dialog window
provides additional control facilities and some information about the cropping
process:
Slice sliders: XY plane/XZ plane/YZ plane Allow to "move" through the
image in Z, Y, and X directions respectively. You can change the slice number
by moving the sliders or by changing the numbers in the spinboxes (type
new number or click on the small arrows).
Old size Shows the original dimensions of the image and the original size of
the image in megabytes.
New size Interactively shows the size of the marked part and its expected size
in megabytes.
MAVI user manual
34
Crop
Figure 14: Crop dialog window.
Physical size Interactively shows the expected physical dimensions of the
marked part in micrometers.
Note that cropping a 3D image to a new image size of 1 in one coordinate
direction still results in a 3D image. For slices in the XY plane conversion to a 2D
image can be achieved by Save Image with “2d Image” as target file type. For
differently oriented slices, the “Save current slice view” option from the
“Settings” toolbox of the view control window of the Slice View can be used.
MAVI user manual
35
AutoCrop
8.2.2
AutoCrop
Images of type can be cropped automatically. Choose a minimum pixel distance
to the image border and a minimal grayvalue. MAVI crops the image such that
each pixel with a grayvalue greater than the given minimal one has at least the
given distance to the border in the cropped image.
8.2.3
Extract by label
This function transforms a gray value image into an image whose foreground
consists of all pixels with the chosen gray value or label in the original image.
The user chooses for the resulting image either the type MONO or to preserve
the original image type. In the first case the foreground pixels get value 1 while
in the latter they keep their original gray value.
In particular, it is possible to extract the background (label 0), too. Extracting the
background from a label image and preserving the image type results in label 1
for the new foreground, see Figure 15 for an example.
(a) Dialog
(b) Before extracting 0
(c) After extracting 0
Figure 15: Extract by label dialog window and effect of extracting label 0 from
a label image.
8.2.4
Pad
Pad embeds the image into a larger image by adding slices at the borders. The
number of slices to be added to each side of the image as well as the gray value
for the new pixels can be specified in the padding dialog.
Note that the dialog will allow to choose the value to embed the image in
within the range of the image type only.
Padding a 2D image can result in a larger 2D image, if rows are added in X and
Y directions, or in a 3D image, if slices in Z direction are added, too.
MAVI user manual
36
Shrink
8.2.5
Shrink
This function reduces the size of an image by virtually decreasing the lateral
resolution. The reduction factor can be set for each coordinate direction
separately. The new image size in the respective coordinate direction is the old
image size divided by the reduction factor.
In the case of a 3D image, the value of a pixel in the resulting image is the
average of all pixel values within a cuboid of size ReductionFactorX x
ReductionFactorY x ReductionFactorZ of the input image. For a 2D image, the
resulting value is the average of all pixel values within a rectangle of size
ReductionFactorX x ReductionFactorY.
8.2.6
Cast
Cast changes the gray value type of the image by simply casting each pixel
value to the target gray value type. When casting down to a smaller gray value
range, this results in a loss of image information since values exceeding the new
gray value range are truncated. For example, when casting to GRAY8, each
pixel with a value smaller than 1 is assigned 0 as new gray value while each
pixel with an original value greater than 255 is assigned 255.
8.2.7
Convert
Converting is a linear scaling of the gray values from one gray value range to
another, where the dynamic range of the gray values remains the same in
relation to the value range. Thus Convert preserves the full image information
and keeps the relations between gray levels as intact as possible, see Figure 17
for a sketch.
Here dynamic range denotes the difference between maximum and minimum
gray value in the image while gray value range refers to the difference between
maximum and minimum possible gray values determined by the image type,
e.g. 255 for GRAY8 images.
8.2.8
Spread
Spread is a simple transformation of the image type. Depending on the user’s
choice, the gray values are transformed linearly, logarithmically, or following the
square or root functions.
Spreading transforms the dynamic range, that is, the range from minimum to
maximum value present in the input image to the full value range of the output
MAVI user manual
37
Equalize
image. Spread preserves all image information, i.e. it does not cut off any
values. It may however transform different gray levels onto the same one, when
spreading down e. g. from GRAY16 to GRAY8 images.
After spreading, the full value range of the resulting image will be used. For
example, when a GRAY8 image is spread to GRAY16, the smallest value in the
original will be assigned a zero, and the largest value in the original image will
be assigned the value 216 − 1.
8.2.9
Equalize
Equalization is a method that can be used to enhance the contrast of an image.
In MAVI, equalization to an output image type diffferent from the input image
type is possible, too.
Equalize uses a transfer function which is derived from the image’s gray value
histogram. Hence, the output of equalization depends on the image content.
Usually, contrast will be enhanced in the resulting image. That is, structures that
were hardly visible before may appear much brighter after equalizing. However,
equalization tends to emphasize noise, too.
Figure 16 shows how Cast, Spread, and Equalize affect a GRAY8 image.
Figure 17 visualizes the difference of Spread and Convert.
(a) Original
(b) Cast
(d) Transfer function
(e) Equalize
(c) Spread, linear
Figure 16: Changing of image types by an example. Shown are the images’ gray
value histograms and the transfer function for Equalize.
MAVI user manual
38
Prepare label image for
visualization
Figure 17: Visualization of the difference of Spread and Convert. Spread always
uses the full target gray value range whereas the converted image occupies the
same proportion of the gray values of its range as the original did.
8.2.10 Prepare label image for visualization
Reduces the number of labels in images obtained by Labeling or Watershed
Transformation to the number N given by the user, in order to facilitate the
View:Volume Rendering View. Values N > 20 are not recommended. See
Section8.6.3 for details.
The labels are assigned in turns and spread to GRAY8. That is, if an object had
label M in the input image, it is assigned the new label
(M mod N ) + 1 ∗ 256/(N + 1).
Note that this is a helper function for visualization purposes only. Do not base
any analysis on the result of this function as different objects are assigned equal
labels.
8.2.11 Complement
This function assigns to each pixel its complementary pixel value, which is
obtained by subtracting the current value from the maximum of the gray value
range given by the image type.
The complement of a MONO image simply swaps fore- and background, see
Figure 18(c) and (d). For an 8 bit gray value image, the largest possible pixel
value is 28 − 1 = 255. A pixel value of 12 will be converted into 255 − 12 = 243
by Complement. See Figure 18(a) and (b) for an example.
MAVI user manual
39
Complex Conjugate
(a) Original
(b) Complement
(c) Binarized original
(d) Complement
Figure 18: Example for Complement acting on gray value or binary (MONO)
images.
8.2.12 Complex Conjugate
This function computes the complex conjugate of a COMPLEX image, that is, it
assigns each pixel with complex value z = x + i ∗ y the new value z = x − i ∗ y.
8.2.13 Mask Image
To mask an image means to select certain areas of the image and discard the
rest as background. The areas to be selected are specified by the foreground of
a second image, called the "mask".
The binary operation Mask Image is active only if a MONO mask image is
currently selected. The image on which the operation will be applied can then
be selected via the dialog. Only images having the same size as the mask image
can be chosen.
A very simple example of Mask Image is shown in Figure 19: The original
MAVI user manual
40
Mask with Cylinder
image is masked by the image having a white rectangle in the foreground. Both
images are of type MONO. The foreground of the result image contains only
those pixels of the original image that were "covered" by the white rectangle of
the mask image. For MONO images Mask Image is equivalent to a pixelwise
logical And.
(a) Original
(b) Mask
(c) Result
Figure 19: Mask Image, simple example.
Figure 20 shows a more realistic example application of Mask Image:
Tomographic reconstruction artefacts are removed from the image background
by applying Binarization followed by Mask Image.
(a) Original
(b) Mask
(c) Result
Figure 20: Mask Image, realistic example.
8.2.14 Mask with Cylinder
In order to facilitate analysis of the full volume for cylindrical samples, MAVI
offers this special function for defining and applying a cylindrical mask. The
MAVI user manual
41
Unary Operations
cylinder’s base is parallel to the coordinate plane given by the currently active
Slice View. The user specifies the
• Cylinder radius [pixel] and
• Cylinder center [pixel]
in the dialog box. Moreover, the user decides whether the reulting image keeps
the size of the original one or is reduced to the bounding box (of the cylinder)
by the New image size parameter.
Figure 21: Mask with Cylinder dialog box.
8.2.15 Unary Operations
Unary operations derive the new image from the original one by pixelwise
calculating the new gray value from the old one. The following operations are
available:
• Add Value
• Subtract Value
• Multiply by Value
• Divide by Value
The second value is given by the user via the dialog.
Unary operations cannot be applied to MONO images. For gray value images,
the pixel values are truncated if the result of the chosen operation exceeds the
MAVI user manual
42
Binary Operations
gray value range defined by the image type.
8.2.16 Binary Operations
A binary operation is a calculation which takes pixel values from two different
images at the same spatial location and writes the resulting value to the output
image before proceeding the next location within both input images. For
example, you can add two images and the value at a location in the resulting
dataset will contain the sum of both values at the corresponding locations in the
two input images.
Images must be of equal type and dimensions. If the resulting pixel value is
greater than the maximum or lower than the minimum of the given value type,
the resulting value is set to the maximum respectively the minimum.
In order to link two images of different gray value type, Cast the one with the
smaller gray value range to the type of the other image.
The following binary operations are available:
Add: pixelwise sum of the gray values, for MONO images equivalent to the
logical operation Or
Subtract: pixelwise difference of the gray values, for MONO images A and B
equivalent to the logical A And Complement(B)
Minimum: pixelwise minimum of gray values, for MONO images equivalent to
the logical operation And
Maximum: pixelwise maximum of the gray values, for MONO images
equivalent to the logical operation Or
And: pixelwise logical and, on GRAY value images pixelwise minimum
Or: pixelwise logical or, on GRAY value images pixelwise maximum
Multiply: pixelwise product of the gray values
Absolute Difference: absolute value of the pixelwise difference, useful for
comparison of images.
MAVI user manual
43
Segmentation
8.3
Segmentation
Segmentation refers to two tasks – finding the component or image segment of
interest in a gray scale image and identifying connected objects or regions. We
call the first task “binarization” as the output image is binary (MONO), i. e. it
contains only zeros and ones as pixel values. Binarization is a prerequisite for
measuring geometric characteristics of structural components like the Field
Features.
Currently available binarization methods are global thresholding, hysteresis
thresholding[1], and local thresholding according to Niblack[8] and Sauvola[16].
Segmentation methods for identifying foreground regions are labeling and the
watershed transformation. Input data for labeling is a MONO image while the
watershed transformation expects gray valued input. The output images contain
discrete label values identifying connected regions. These are needed to
compute geometric features of objects like the Object Features.
Integrated plugins combine several single functions to offer complex
functionality at one click. For separation of touching particles like the grains of a
powder or for the reconstruction of cells of an open cell foam, Labeling will fail
as the objects are connected in the binary image. A combination of the
Euclidean Distance Transformation with the Watershed Transformation
will however do the job, provided n be prevented without destroying the
structural information [12, Section 4.2.6]. An example is given in tutorial 7.1.2.
In most analyses, segmentation is performed rather early within the processing
chain. For very noisy images, some sort of smoothing, usually
Transformation:Filtering, should be applied before segmenting.
8.3.1
Binarization
Global thresholding is the simplest binarization method and based on the idea,
that the component of interest can be identified solely based on its brightness.
This is for instance surely fulfilled in the case of the CT image of a porous
structure.
The binary image with values g is created from the original image with gray
values f by comparing the value of each pixel with the lower and the upper
threshold values:
1 if lowerthreshold ≤ f (x) ≤ upperthreshold
g(x) =
0 otherwise.
The thresholds have to be determined either interactively by the user, usually
with the help of the gray value histogram of the image and visual feedback, or
MAVI user manual
44
Hysteresis thresholding
by a threshold selection scheme. Slicewise both histogram and preview are
available within the Binarization dialog window, see Figure 8.3.1. Checking the
Whole Image Histogram box induces MAVI to calculate the gray value
histogram of the whole image. This also activates the Otsu’s threshold button to
be used to let MAVI suggest a lower global threshold according to Otsu’s
method. Otsu’s threshold maximizes the variance of gray values between the
two classes created while keeping the variance within the classes as small as
possible:
2
Mean[{f (x)|f (x) < T }] − Mean[{f (x)|f (x) ≥ T }]
TOtsu = argmaxT
.
Variance[{f (x)|f (x) < T }] + Variance[{f (x)|f (x) ≥ T }]
(1)
Step-by-step, the binarization works as follows:
(1) Use the slider Slice number until a representative slice through the image
is shown.
(2) The histogram of the current slice is displayed in the Information box.
Switch to the histogram of the whole dataset by checking Whole Image
Histogram.
(3) Switch to Logarithmic scale for the gray value-axis in the Histogram plot
by pushing the respective button, in case the gray value histogram is hard
to read due to the huge number of dark pixels.
(4) Use the sliders to set Lower threshold and Upper threshold, respectively.
The result is immediately visible in the Preview box.
(5) Adjust the sliders until the slice preview is satisfying.
(6) Verify that the selected thresholds are suitable for the other slices using
the Slice number slider again.
(7) Click OK to apply the selected thresholds to the whole dataset.
Remark 3 The interval defined by upper and lower threshold is half open,
including the lower but excluding the upper one. That is, to select exactly the
pixels with value e.g. 128 you need to choose lower threshold 128 and upper
129.
8.3.2
Hysteresis thresholding
Hysteresis or double thresholding [1] transforms a gray value image into a
MONO image by a two-step procedure. A lower and an upper threshold value
Tl < Tu are input. First a simple global thresholding (1) with Tu as lower and the
maximal gray value as upper threshold is applied. Subsequently, the foreground
is enlarged according to the following rule:
For each pixel neighboring a foreground pixel, it is checked whether its value
exceeds the second (lower) threshold Tl . The new value of the current pixel is
one if this condition is fulfilled and zero otherwise. This is repeated as long as
MAVI user manual
45
Niblack Segmentation
the foreground still grows, see Figure 23 for a sketch of the algorithm.
Hysteresis thresholding yields smooth connected foreground components
without emphasizing noise.
8.3.3
Niblack Segmentation
Global thresholding fails whenever noise is strong compared to the image
content or when the image features global gray value fluctuations as due for
instance to a strongly absorbing material in CT or to the tilted electron beam in
FIB-tomography. Local thresholding can overcome this disadvantage of global
thresholding by spatially varying the gray value threshold t(x):
1
if f (x) ≥ t(x)
(2)
g(x) =
0
otherwise.
Niblack’s algorithm [8] deduces the local threshold t(x) for the current pixel x
from gray value mean and variance within a cubic window W (x) centered at x:
(
q
1
if f (x) ≥ MeanW (x) (f (x)) + c VarianceW (x) (f (x))
g(x) =
(3)
0
otherwise.
The window size (edge length) is a parameter of the method and should be
chosen according to the size of objects or the structure. The choice of the
parameter c depends on the noise. In [21] c = −0.2 was recommended.
However, that study was restricted to character recognition and thus it can
hardly be generalized. In most cases the parameters have to be determined by
trial-and-error, which is time consuming and should therefore be tried on a
small but representative sub-volume created by Manipulation:Crop first.
In the dialog shown in Figure 24, the user has to specify the Window Size
corresponding to the number of pixels to inspect in each direction from the
current pixel. Thus, the resulting cubic window has the side lengths 2 *
Window Size + 1. This parameter should be adjusted to the approximate size of
the image features to be segmented. The second parameter Coefficient is the c
from (3) and governs over- and under-segmentation, see Figure 25.
In the current version, no edge treatment is implemented. Thus, the resulting
MONO image will have a black border exactly Window Size pixels wide.
Local binarization is available for all image gray value types including FLOAT.
However, for FLOAT images, the option Use fast implementation must be ticked.
Note that the result of Cast to FLOAT followed by Niblack Segmentation can
differ from the local binarization of the original GRAY8 or GRAY16 image in
some pixel values due to rounding effects.
MAVI user manual
46
Sauvola Segmentation
8.3.4
Sauvola Segmentation
(
g(x) =
8.3.5
1
0
In images with large low contrast regions, Niblack’s local thresholding tends to
emphasize noise or low frequency variations due to the low local gray value
variance. An empirical modification by J. Sauvola and M. Pietikäinen [16] solves
this problem:
q
if f (x) ≥ MeanW (x) (f (x)) + c MeanW (x) (f (x)) R1 VarianceW (x) (f (x)) − 1
otherwise.
(4)
The additional parameter R normalizes the variance. In MAVI it is always set to
half of the image’s gray value range, that is R = 128 for GRAY8 and
R = 215 = 32 768 for GRAY16 images. For FLOAT images f , R is set to
(max f − min f )/2. For negative Coefficients c, the local threshold is now close
to the local mean in high contrast areas and significantly above the local mean
in low contrast areas.
Labeling
Labeling assigns to every connected object in a MONO image a certain gray
value which is unique for each object. By default, the image foreground is
labeled. However, the user can switch to labeling the background by un-ticking
the corresponding box. MAVI’s labeling is based on Labeling by Reconstruction
from [25]. An example is shown in Figure 26.
The result of the labeling might depend on the chosen adjacency system, see
Section 4.3 for the list of options. If the structure is sufficiently well resolved and
smooth, it does not matter which adjacency system is assumed. However, if one
desires to keep thin, not well resolved structures connected, the 26 adjacency
system should be used in 3D (8 in 2D), while the 6 and 4 adjacency systems are
preferrable when touching object are to be separated.
The number of objects found is visible in the Info tab of the View Control
Window and stored in the image header’s history entry. The gray value type of
the labeled image is automatically adapted to the number of objects found.
8.3.6
Watershed Transformation
The watershed transformation segments a gray value image by simulation of
flooding the image starting at the smallest gray values following [24]. Where
waters from two sources would collate, a watershed is set. Watersheds get label
0, while each basin is assigned a specific label. The watersheds separate the
basins. That is, the basins are not connected with respect to the chosen
adjacency system. An example is shown in Figure 27.
MAVI user manual
47
Watershed Transformation,
Preflooded
The watershed transformation is an essential ingredient for cell reconstruction
as performed by the integrated plugins for this task as well as in the
step-by-step tutorials for the open nickel foam and the closed zinc foam. In the
same way, the watershed transformation is used by the integrated plugin for
particle separation.
Usually, when applied to gray value images, the waterhed transformation results
in strongly over-segmented images. A way to avoid over-segmentation is to use
the preflooded version instead. For other strategies see the integrated plugins
for cell reconstruction.
8.3.7
Watershed Transformation, Preflooded
Preflooding is a method to avoid over-segmentation in the Watershed
Transformation. When flooding the image, at each step the number of pixels
in the created basins is checked. If it is below the minimal pixel number
specified as Area Threshold in the dialog, the basin is blocked from creating
watersheds. It will be integrated into a larger basin instead. An example is
shown in Figure 28.
This procedure is equivalent to
(1) flooding the image until the basins attain the minimal pixel number,
(2) assigning all pixels associated to a specific basin the highest gray value
present in that basin, and
(3) using the resulting image as input for the Watershed Transformation.
MAVI user manual
48
Watershed Transformation,
Preflooded
Figure 22: Binarization dialog window.
MAVI user manual
49
Watershed Transformation,
Preflooded
Figure 23: The principle of hysteresis thresholding illustrated by a 1d example:
dark gray - foreground due to pixel value higher than higher threshold Tu , light
gray - foreground due to being connected to dark gray area and pixel value
higher than lower threshold Tl , white - background.
Figure 24: Niblack Segmentation dialog window.
MAVI user manual
50
Watershed Transformation,
Preflooded
(a) Original
(b) Over-segmentation
(c) Correct
(d) Under-segmentation
Figure 25: Niblack over- and under-segmentation. For Coefficient=-0.5, oversegmentation occurs: Areas which clearly belong to the background are assigned
to the foreground. For Coefficient=-0.1, a good segmentation result is achieved.
For Coefficient=0.1, under-segmentation occurs: Areas clearly belonging to the
foreground are assigned to the background.
MAVI user manual
51
Watershed Transformation,
Preflooded
(a) MONO input
(b) Labeled result
Figure 26: Labeling example.
(a) Original
(b) Watershed result
Figure 27: Watershed Transformation example.
MAVI user manual
52
Watershed Transformation,
Preflooded
(a) Input
(b) Watershed result
(c) Watershed result
Figure 28: Effect of Watershed Transformation, Preflooded: (b) was obtained
from (a) by a simple Watershed Transformation. Over-segmentation is clearly
visible in (b) but can be avoided by preflooding as can be seen in (c).
MAVI user manual
53
Cell Reconstruction
8.3.8
Cell Reconstruction
In the reconstruction of cells, different methods can be used to prevent overand undersegmentation. Cell Reconstruction by Smoothing uses the
Binomial Filter to smooth the distance image before the Watershed
Transformation. The only parameter to be set is the size of the filtermask. Cell
Reconstruction by Complex Morphology uses the H-Maxima option of the
H-Extrema transformation to suppress regional maxima. In the dialog, the s to
be set. Finally, Cell Reconstruction by Preflooded Watershed exchanges the
Watershed Transformation for the preflooded one. It takes as input
parameter the minimal pixel number that all regions created by the watershed
transform have to contain.
Cell Reconstruction by Smoothing is the fastest choice. However, this simple
gray value smoothing works rarely.
Cell Reconstruction by Preflooded Watershed works much better. A good
way to choose the parameter “Area Threshold” is either to exploit knowledge
about the material to find a lower bound for the cell volume or to use the mean
cell volume from the Open Foam Features. Divide by the volume of one pixel
(s1 s2 s3 , where (s1 , s2 , s3 ) is the spacing vector) to get the mean cell volume in
[pixel]. Depending on the structure, use 10 − 40% of that value. The only
drawback of the Preflooded Watershed Transformation is, that it slows down
considerably for large regions and a high “Area Threshold”.
In that case, Cell Reconstruction by Complex Morphology is the best
alternative. Its parameter “DynamicH” is however harder to choose.
“DynamicH” is related to the diameter of the cells.
To get a rough estimate for the parameter for open foams use the mean cell
diameter from the Open Foam Features d. Divide by the spacing d/s1 as the
dynamic is given in pixel units. Estimate visually or using the mouse slider in the
Slice View the radius of the largest cell R (in [pixel]). This value is mapped to
the value 255 by the Spread following the Euclidean Distance
Transformation. Thus a starting value for the dynamic is 10 − 40% of
255 · d/(R · s1 ).
For closed cell structures, the mean chord length MIL obtained by the Field
Features on the Complement can be exploited. Again, it has to be divided by
the spacing and the effect of the Spread has to be taken into account. A
starting value for the dynamic is now 1 − 2% of 255 · MIL/(R · s1 ).
MAVI user manual
54
Particle Separation
8.3.9
Particle Separation
The separation of touching particles uses the same algorithmic ingredients as
the Cell Reconstruction. The major difference is, that the Euclidean Distance
Transformation is applied to the foreground of the input MONO image instead
of the background.
Particle Separation by Smoothing uses the Binomial Filter to smooth the
distance image before the Watershed Transformation. The only parameter to
be set is the size of the filtermask.
Particle Separation by Complex Morphology uses the H-Maxima option of
H-Extrema to suppress regional maxima. The dynamic is user defined. See
8.3.8 for details on the choice of the parameter.
Particle Separation by Preflooded Watershed uses the Preflooded
Watershed Transformation to prevent oversegmentation. It takes as input
parameter the minimal number of pixels that a particle has to contain. See 8.3.8
for details on the choice of the parameter.
Tutorial 7.4 describes the separation of sintered copper particles step-by-step.
MAVI user manual
55
Transformation
8.4
Transformation
8.4.1
Filter
Filters replace the value of a pixel by combining the pixel values in a neighboring
region given by a filter mask or structuring element. The filter function can be a
linear combination of input values (linear filter) or a more general non-linear
function.
Filtering is typically used in the initial steps of the processing chain for denoising,
smoothing, and contrast enhancement to facilitate segmentation or analysis at
later stages.
Smoothing filters help to denoise images, often at the cost of blurring edges,
whereas edge detection filters are used to emphasize edges of an object or
within complex structures.
The filter masks are all cubic in pixels, no matter how the image spacings are
set. Table 6 lists the filters provided by MAVI.
A Morphological Mean filter with larger and more flexible filter masks is
available in the Morphology submenu.
Laplace Filter 2 is a modifies edge detection filter setting former background
pixels to 0, edge pixels to the largest possible value, and foreground pixels to a
gray value halfway in between.
Rank order filters sort the N 2 or N 3 values of the pixels covered by the filter
mask by size and assign the value at the desired rank from this list to the pixel
considered. The median is the value at the central position N 2 /2 + 1 or
N 3 /2 + 1 for 2D and 3D, respectively.
8.4.2
Morphology
Morphological transformations modify the image content with respect to a
given structuring element. The structuring elements (SE) available in MAVI’s
Morphology submenu are listed in Tables 7 and 9 for 3D and 2D images,
respectively. The structuring element is centered at its central pixel or in the
lower left front one of a set of central pixels. Note that this can cause shifting,
for instance when a cube of even edge length is used repeatedly. For an
isotropic lattice, the SE Approx. Ball is a linear combination (Minkowski addition)
of line
√segments of length Size parallel to the 3 coordinate directions, length
Size/ 2 in the 6 directions
defined by the face diagonals of the lattice unit cell,
√
and length Size// 3 in the 4 directions defined by the space diagonals of the
MAVI user manual
56
Morphology
Filter
Smoothing filters
Mean Filter
Mask size
Mask values or function
mean of all values
Weighted Mean Filter 1
3 ≤ N ≤ 31,
odd
3 ≤ N ≤ 31,
odd
3
Weighted Mean Filter 2
3
Gauss Filter
3
Edge detection filters
High Pass Filter
Laplace Filter 1
3
3
Laplace Filter 2
3
Nonlinear Filters
Alpha-Trimmed Mean Filter
3
Median Filter
3, 5, 7
Binomial Filter
−1
convolution of 1D filters with entries NN−1−k
at distance k
3(N
−1)
2(N
−1)
from center, normalized by 2
(2
)
center = 2, 6(4)-neighbors = 1, all others = 0, normalized by
8(6)
center = 4(3), 6(4)-neighbors = 2, all others = 1, normalized
by 36(15)
center = 4, 6(4)-neighbors = 1(2), all others = 0(1), normalized by 10(16)
center = 26(8), all others = -1, normalized by 26(8)
center = 6(4), 6(4)-neighbors = -1, all others = 0, normalized
by 1
center = 7(5), 6(4)-neighbors = -1, all others = 0, normalized
by 2
mean after removing top and bottom alpha portion from
sorted value list, alpha=0 – mean, alpha=0.5 – median filter
median
Table 6: Filters in MAVI with their masks, values, and parameters. Mask size is
the edge length of the cubic/square filter mask in pixels. The values in braces
refer to the 2D version if this differs from 3D.
lattice unit cell. The diameter of the resulting approximate ball is therefore
considerably larger than Size, see Table 8 for the approximate diameters. For
morphological transformations with balls of intermediate diameters, please use
alternatively the Euclidean Distance Transformation. Note that Approx. Ball
of Size 2 is nothing but the cube of Size 2.
Structuring elements can be chosen via the dialog box shown in Figure 29
Erosion The value of the pixel covered by the central pixel of the structuring
element is set to the minimum gray value of all pixels covered by the structuring
element mask. Erosion can be used to remove small objects, to reduce edge
deformation, or to break thin connections between objects.
For MONO images Erosion reduces the number of foreground pixels and thus
object size or volume of the foreground. For gray value images Erosion is a
minimum filter and thus the image becomes darker. See Figure 30 for an
MAVI user manual
57
Morphology
Name
Cube
Cuboid
Approx. Ball
Line
Custom
Description
cube parallel to the coordinate axes
cuboid parallel to the coordinate axes
rough approximation of a ball
line parallel to one of the 13 discrete directions given by the
cuboidal lattice
Minkowski sum of segments in the 13 discrete directions
given by the cuboidal lattice
Parameters
side length Size
side lengths SizeX, SizeY, SizeZ
Size
length Size
lengths of the segments
Table 7: Structuring elements for 3D morphology. All size parameters are in
[pixel] units.
Size
Diameter
2
2.31
3
11.00
4
16.26
5
25.00
6
26.56
7
35.22
8
44.00
9
45.61
Table 8: Size parameter and corresponding approximate diameter of the SE
Approx. Ball provided for 3D morphological transformations in the case of an
isotropic lattice.
example.
Dilation The value of the pixel covered by the central pixel of the structuring
element is set to the maximum gray value of all pixels covered by the structuring
element mask. Dilation can be used to remove small disturbances (i.e. holes)
within the objects or to connect interrupted objects.
For MONO images Dilation increases the number of foreground pixels and thus
object size or volume of the foreground. For gray value images Dilation is a
maximum filter and thus the image becomes brighter. See Figure 30 for an
example.
Name
Square
Rectangle
Line
Custom
Description
square parallel to the coordinate axes
rectangle parallel to the coordinate axes
line parallel to one of the 4 discrete directions given by the
rectangular lattice
Minkowski sum of segments in the 4 discrete directions given
by the rectangular lattice
Parameters
side length Size
side lengths SizeX, SizeY
length Size
lengths of the segments
Table 9: Structuring elements for 2D morphology. All size parameters are in
[pixel] units.
MAVI user manual
58
Morphology
Figure 29: Dialog box for defining a structuring element for morphological transformations.
Opening Opening is an Erosion followed by a Dilation. It can be used to
remove small objects, to reduce edge deformation or to interrupt thin
connections between objects while keeping the remainder of the image
unchanged.
In the case of gray value images, Opening reduces noise and enhances image
contrast. See Figure 31 for an example.
Closure Closure is a Dilation followed by an Erosion. It can be used to
remove small disturbances (i.e. holes) within the objects or to connect
interrupted objects while keeping the remainder of the image unchanged. Gray
value images appear “blurred” after a Closure. See Figure 31 for an example.
White Top Hat Subtracts the opened image from the original one. White Top
Hat can be used to remove large objects from the image.
To determine an adequate structuring element try opening the image with an
element that removes the small objects completely.
Black Top Hat Subtracts the original image from the closed one. Black Top
Hat can be used to emphasize holes within the objects or small background
areas in the image.
Algebraic Opening The image is opened with respect to line segments in all
the discrete directions as structuring element. The union of all opening results is
MAVI user manual
59
Morphology
(a) MONO input
(b) Erosion
(c) Dilation
(d) Original
(e) Erosion
(f) Dilation
Figure 30: Erosion and Dilation examples. The structuring element Approx. Ball
of Size 3 was used in all cases.
the algebraic opening. For gray value images this translates to the pixelwise
maximum of the opening results.
The SE lengths are controled by the Line parameter, input in [pixel]: In the case
of an isotropic lattice, the line segments are approximately Line pixels long. In
the case of an anisotropic lattice, the segment lengths are approximately
Line*(average of spacings).
Algebraic Closure The image is closed with respect to line segments in all the
discrete directions as structuring element. The intersection of all closing results
is the algebraic closure. For gray value images this translates to the pixelwise
minimum of the closing results.
The SE lengths are controled by the Line parameter, input in [pixel]: In the case
of an isotropic lattice, the line segments are approximately Line pixels long. In
the case of an anisotropic lattice, the segment lengths are approximately
Line*(average of spacings).
MAVI user manual
60
Morphology
(a) MONO input
(b) Opening
(c) Closure
(d) Original
(e) Opening
(f) Closure
Figure 31: Opening and Closure examples. The structuring element Approx.
Ball of Size 3 was used in all cases.
Morphological Gradient The eroded image is subtracted from the dilated
image. The morphological gradient emphasizes edges of a thickness given by
the structuring element.
Morphological Mean The morphological mean acts essentially like the
common Mean Filter. It is however a generalization in the sense that it
averages the gray values of all pixels covered by the structuring element instead
of just using a cubic filter mask. Besides being more flexible, the morphological
mean filter allows for filter mask sizes up to 1283 pixels. Moreover, it is much
faster for large filter masks.
Toggle Mapping Erosion/Dilation Toggle mapping [19] by erosion/dilation
or the morphological shock filter [13] enhances edge contrast by applying
pixelwise Erosion or Dilation. The actual transformation applied at the current
pixel is chosen such that the result differs the least from the original input
MAVI user manual
61
Distance Transformation
image.
Toggle Mapping Opening/Closure Toggle mapping [19] by opening/closure
enhances image contrast by applying pixelwise Opening or Closure. The actual
transformation applied at the current pixel is chosen such that the result differs
the least from the original input image.
8.4.3
Distance Transformation
Distance transformations usually assign each background pixel in a MONO
image the distance to the foreground. The distance information can be used to
speed up successive morphological Dilations with balls of varying radius as
structuring elements: Taking a global threshold in the distance image is
equivalent to a dilation by a ball (w.r.t. the chosen metric).
Moreover, the Euclidean distance transformation is an essential ingredient for
Cell Reconstruction and Particle Separation where it serves as input for the
Watershed Transformation.
The Euclidean Distance Transformation is not just a valuable processing tool
but a means for analysis in its own right as it yields the spherical contact
distribution [20] and can be used to investigate correlations in the spatial
arrangement of components, see [14].
Euclidean Distance Transformation This function computes the exact or
squared Euclidean distance of each background pixel to the closest foreground
pixel or of each foreground pixel to the closest background pixel of a MONO
image. For an example see Figure 33.
The dialog offers the choice of computation on background or foreground
pixels, the choice of an edge treatment, and whether squared rather than exact
Euclidean distance values are stored in the result image.
Normally, the algorithm results in a FLOAT image. If the input image contains no
spacing information or all spacing entries are set either to 0 or 1, the type of
output image can be specified additionally: If GRAY32 is chosen, then the result
image contains squared rather than exact Euclidean distance values. If FLOAT is
chosen, then the result can be either the squared or exact Euclidean distance
image.
This implementation will be replaced by the Euclidean Distance
Transformation (Voronoi algorithm).
MAVI user manual
62
Distance Transformation
Euclidean Distance Transformation (Voronoi algorithm) This alternative
implementation to Euclidean Distance Transformation based on [7] provides
faster computation of the Euclidean distance image. The choice of edge
treatments is however currently restricted to NET and EMB.
Discrete Distance Transformation This function computes for each
background pixel in a MONO image its distance to the closest foreground pixel
w.r.t. the chosen discrete metric. See Figure 32 for the choices and Figure 33 for
an example.
(a) Manhattan
(b) Maximum
(c) Euclidean
Figure 32: Discrete metrics visualized by equal radius spheres. Manhattan or city
block metric: shortest paths with steps only in the directions of the coordinate
axes. Maximum metric: shortest paths with steps only in the coordinate directions. The discrete Euclidean sphere is shown for comparison.
In analogy to the discrete Manhattan and maximum metrics, MAVI can also
measure discrete distances with respect to the 14.1 and 14.2 adjacency systems.
That is, MAVI measures the length of the shortest discrete path from a
background pixel to the closest foreground pixel that is connected with respect
to the given neighborhood.
MAVI user manual
63
Distance Transformation
(a) Original
(b) Manhattan
(c) Maximum
(d) 14.1
(e) 14.2
(f) Euclidean
Figure 33: Distance transformation results for various choices of the metric. The
brighter the pixels the higher the distance values.
MAVI user manual
64
Geodesic Transformation
8.4.4
Geodesic Transformation
In contrast to the morphological operations operating on an input image
together with a certain structuring element, geodesic transformations use two
input images of the same size. The basic procedure is as follows: a
morphological operation, that is, erosion or dilation by the elementary isotropic
structuring element, is performed on the first image and it is then forced to
remain either above or below the second image. New algorithms are created by
choice of suitable pairs of input images and iterated application of this basic
procedure.
All algorithms in this submenu are based on [19].
Regional Extrema A regional minimum (maximum) of a GRAY8 or GRAY16
image is a connected component of pixels of constant gray value such that any
path to a pixel of a lower (higher) gray value includes at least one pixel of a
higher (lower) gray value.
The algorithm computes either Regional Minima or Regional Maxima. The
adjacency system defining the connectivity of the extrema components is
chosen by the user, too.
The resulting MONO image has the regional minimum (maximum) pixels as
foreground and all the other pixels as background.
H-Extrema The H-Extrema function computes regional extrema of an image
under a contrast constraint and modifies the image by use of these regional
extrema.
An important concept in the context of regional extrema is that of depth or .
The dynamic of an image minimum (maximum) is the minimal height one has to
ascend (descend) to get from the given minimum (maximum) to another
minimum (maximum) of lower (higher) gray value. The dynamic of the global
minimum (maximum) of the image is just the difference between maximal and
minimal gray value in the image.
The H-Minima (H-Maxima) transformation reduces the number of regional
minima (maxima) of an image by use of a contrast criterion. It suppresses all
minima (maxima) whose dynamic is smaller than the threshold Dynamic to be
set by the user. This transformation is well suited to prevent oversegmentation
in the Watershed Transformation and one option for smoothing in the Cell
Reconstruction and Particle Separation algorithms.
MAVI user manual
65
Geodesic Transformation
The Extended H-Minima (Extended H-Maxima) extracts connected regions
surrounding the local minima (maxima). It is equivalent to applying Regional
Extrema to the result of H-Minima (H-Maxima). In the MONO output image, all
pixels belonging to an H-Minimum (H-Maximum) are set to foreground, while
all the other pixels are set to background.
The H-Concave option subtracts the H-Minima transformed image from the
input image. The H-Convex transformation is the difference of the input image
and its H-Maxima.
Adaptive H-Extrema The Adaptive H-Extrema compute regional extrema
of an image under a gray value dependent contrast constraint and modify the
image by use of the computed regional extrema.
Adaptive H-Extrema is a modified version of H-Extrema with a different
contrast criterion. More precisely, the Adaptive H-Extrema adapt the Dynamic
parameter to the total height in the gray value relief, whereas in the H-Extrema
transformations the Dynamic is a constant independent of the actual gray value
of the extremum currently considered.
The dialog offers the same choice of transformations as for the H-Extrema, the
Neighborhood, and two parameters for the dynamic: the Dynamic at 0 and the
Dynamic at maximum, where maximum is the largest possible value in the range
of the image type. The dynamic is interpolated linearly between these two
values. This is illustrated in Figure 34. Adaptive H-Minima can save the
separation of foreground or background objects if Particle Separation and
Cell Reconstruction fail due to strongly diverging sizes of the particles or
pores, respectively. For an example of a practical application see the Tutorial 7.2.
Minima Imposition This function helps to remove irrelevant minima as e. g.
due to noise. It requires two input images: a MONO marker image marking
relevant dark image regions and an image on which the minima are imposed. A
suitable marker image can be obtained from the second image by an
appropriate transformation such as Regional Extrema.
The Minima Imposition preserves or creates image minima at the locations
indicated by the markers and removes those not corresponding to a marker. The
MONO marker image has to be selected while the second image is chosen via
the dialog.
Minima Imposition is useful for preprocessing to prevent oversegmentation by
the Watershed Transformation, provided the marker image contains a
nucleus of every segment. See Figure 35 for an example.
MAVI user manual
66
Geodesic Transformation
Figure 34: Calculation of dynamic h in Adaptive H-Extrema.
Maxima Imposition This function helps to remove irrelevant maxima as e. g.
due to noise. It requires two input images: a MONO marker image marking
relevant bright image regions and an image on which the maxima are imposed.
A suitable marker image can be obtained from the second image by an
appropriate transformation such as Regional Extrema.
The Maxima Imposition preserves or creates image maxima at the locations
indicated by the markers and removes those not corresponding to a marker. The
MONO marker image has to be selected while the second image is chosen via
the dialog.
Ultimate Eroded Set This transform reduces the connected components of
the foreground of the MONO input image to their central regions.
The Ultimate Eroded Set is generated by successive erosions of the input
image (type MONO) using the structuring element given by the chosen
neighborhood. Each time an erosion step would remove a connected
component, the remainder of this connected component is kept. That is, the
Ultimate Eroded Set is the union of all last pixels of connected components
that disappear during successive erosions of the input image. The ultimate
eroded set coincides with the regional maxima of the Discrete Distance
Transformation on the foreground.
MAVI user manual
67
Geodesic Transformation
(a) Marker
(b) Original
(c) Minima imposed
(d) Watershed on (b)
(e) Watershed on (c)
(f) Mask Image (e) with Binarization
Figure 35: Minima Imposition used as preprocessing for Watershed Transformation.
Ultimate Dilated Set This transform finds the central regions of the
connected components of the background of the MONO input image.
The Ultimate Dilated Set is generated by successive dilations of the input
image (type MONO) using the structuring element given by the chosen
neighborhood. Each time a dilation step would remove a connected component
of the background, the remainder of this connected component is kept. That is,
the Ultimate Dilated Set is the union of all first pixels of background
connected components that appear during successive dilations of the input
image. The ultimate dilated set coincides with the regional maxima of the
Discrete Distance Transformation on the background.
Reconstruction by Erosion A reconstruction by erosion is a special form of a
geodesic erosion and involves two input images: A mask image and a marker
image. Both images must be of equal size and the mask image must be
pointwise less than or equal to the marker image.
MAVI user manual
68
Geodesic Transformation
The geodesic erosion of the marker image with respect to the mask image is
obtained as follows: First, the marker image is eroded and then the pointwise
maximum of the resulting eroded image and the mask image is taken. Thus, the
resulting image is forced to remain above the mask image. The iteration of this
basic procedure always reaches stability, in the sense that the resulting image
can no longer be modified, after a finite number of steps.
The reconstruction by erosion of a mask image from a marker image is defined
as an iteration of geodesic erosion of the marker image with respect to the
mask image, stopped as soon as stability is reached. To compute the
Reconstruction by Erosion in MAVI, the marker image has to be selected.
The dialog helps in choosing the mask image, which has to be of the same type
and pixelwise less than or equal to the marker image. The reconstructed image
is pointwise greater than or equal to the mask image and less than or equal to
the marker image.
Reconstruction by Dilation A reconstruction by dilation is a special form of a
geodesic dilation and involves two input images: A mask image and a marker
image. Both images must be of equal size and the mask image must be
pointwise greater than or equal to the marker image.
The geodesic dilation of the marker image with respect to the mask image is
obtained as follows: First, the marker image is dilated and then the pointwise
minimum of the resulting dilated image and the mask image is taken. Thus, the
resulting image is forced to remain below the mask image. The iteration of this
basic procedure always reaches stability, in the sense that the resulting image
can no longer be modified, after a finite number of steps. The reconstruction by
dilation of a mask image from a marker image is defined as an iteration of
geodesic dilations of the marker image with respect to the mask image, stopped
as soon as stability is reached, see Figure 36. To compute the Reconstruction
by Dilation in MAVI, the marker image has to be selected. The dialog helps in
choosing the mask image, which has to be of the same type and pixelwise
greater than or equal to the marker image. The reconstructed image is
pointwise less than or equal to the mask image and greater than or equal to the
marker image.
Self Dual Reconstruction This function reconstructs the marker image f by
self dual geodesic transformation with respect to the mask image g. Depending
on the local relation of the pixel values in marker and mask image, the pixel is
assigned the value of the Reconstruction by Erosion or the Reconstruction
by Dilation.
MAVI user manual
69
Geodesic Transformation
The self dual reconstruction is defined by:
d R (f ) (x), iff (x) ≤ g(x),
[Rg (f )] (x) = ge Rg (f ) (x), otherwise
Here Rgd (f ) denotes the reconstruction by dilation of f w.r.t. g and Rge (f ) the
reconstruction by erosion of f w.r.t. g.
Self-dual reconstruction can ease segmentation, in particular in cases where
contrast and noise are of the same magnitude. To achieve this, use a strongly
smoothed version of the original image f as mask image g as provided by the
Morphological Mean filter or an Opening.
MAVI user manual
70
Geodesic Transformation
(a) Marker and mask
(b) Dilation
(c) Geodesic dilation
(d) Geodesic dilation 2
(e) Geodesic dilation 3
(f) Geodesic dilation 4
(g) Geodesic dilation 5
(h) Geodesic dilation 6
(i) Reconstructed image
Figure 36: Reconstruction by Dilation illustrated by a 1D example. The shaded
image in (i) is the resulting reconstructed image.
MAVI user manual
71
Spectral Transformation
8.4.5
Spectral Transformation
This submenu provides a fast Fourier transform for 3D images as well as it’s
inverse. The fast Fourier transform allows to determine second order
characteristics describing fluctuations of microstructures like the
Autocovariance fast and efficiently.
Fast Fourier Transformation Fast Fourier transformation of the input image.
Uses FFTW 3.0.1 [2]. Results in a COMPLEX image.
Inverse Fast Fourier Transformation Inverse of the Fast Fourier
Transformation. Operates on input images of type COMPLEX. Uses FFTW
3.0.1 [2]. Results in a GRAYF image.
Power Spectrum Power spectrum of input image. Uses FFTW 3.0.1 [2].
Results in a FLOAT image. Use Power Spectrum (Logarithmic) to obtain a result
in logarithmic scale.
Magnitude Magnitude of Fast Fourier Transformation of the input image.
Uses FFTW 3.0.1 [2]. Results in a GRAYF image. Use Magnitude (Logarithmic) to
obtain a result in logarithmic scale.
Phase Phase of Fast Fourier Transformation of the input image. Uses FFTW
3.0.1 [2]. Results in a GRAYF image. Use Phase (Logarithmic) to obtain a result
in logarithmic scale.
MAVI user manual
72
Cut Hills
8.4.6
Cut Hills
The hills of an image are its regional maxima not connected to the image
border. The Cut Hills transform removes them by setting them to the lowest
gray value they are connected to.
The actual hills of the image can be obtained using Extract Hills or by
subtracting the transformed image from the original one.
8.4.7
Fill Holes
The holes of an image are its regional minima not connected to the image
border. The Fill Holes transform removes them by setting them to the highest
gray value they are connected with.
The actual holes of the image can be obtained using Extract Holes or by
subtracting the original image from the transformed one.
8.4.8
Extract Hills
The hills of an image are its regional maxima not connected to the image
border. Extract Hills deduces the hills of the image by subtracting the Cut Hills
transformed image from the original one.
8.4.9
Extract Holes
The holes of an image are its regional minima not connected to the image
border. Extract Holes deduces the holes of the image by subtracting the original
image from the Fill Holes transformed one.
8.4.10 Skeleton
The skeleton of an object in a MONO image is obtained by thinning the object
while preserving its topology. It is a connected set of faces and branches and
contains important information about the geometric structure of the object.
The current algorithm is a sequential version of [22] and has been adapted to
the other adjacency systems. In the Skeleton dialog, the following parameters
can be specified:
Neighborhood adjacency systems w. r. t. which the topology is preserved,
Skeleton Type can be either medial surface, in which case the skeleton will be
a connected set of surfaces, or medial axis,
MAVI user manual
73
Skeleton Analyzer
Branching factor is used to control the preservation of end points.
8.4.11 Skeleton Analyzer
The Skeleton Analyzer assigns every Skeleton pixel a gray value representing
the point class it belongs to. The classes, their labels, and the corresponding
color used in the Slice View are listed in Table 10.
Label
Point class
1
isolated point
2
inner point of branch
3
branch end point
4
junction point of branches
5
junction point of face and branch
6
inner point of face
7
edge point of face
8
junction point of faces
Color
Table 10: Classes of Skeleton points specified by Skeleton Analyzer and corresponding colors used for their representation.
8.4.12 Object Filter
The Object Filter allows to remove objects from a label image as generated by
Labeling or Cell Reconstruction from the Segmentation menu. The removal
decision is based on one, or a combination of, the following three
characteristics of an object: its volume, its shape factor f1 (sphericity), and
whether or not the object touches the image border. Thus calculation of the
Object Features is a prerequisite for applying the Object Filter to obtain the
geometric information for the objects.
In the Object Filter dialog window, use the slider Show slice to view a
representative cross section. Choose the removal criterion to be applied by
clicking on the appropriate tab, then adjust the parameters as described in the
following. The result is immediately visible in the preview box labeled Filtered.
For Volume: Use the sliders labeled Minimum Volume and Maximum Volume to
set lower and upper volume bounds, respectively. For Border: Select only border
objects and only non-border objects to keep the specified objects. For
MAVI user manual
74
Map Subfield Feature To Image
Sphericity: Use the Minimal Shape Factor Sphericity and Maximal Shape Factor
Sphericity to set lower and upper boundaries for the shape factor.
Adjust the parameter settings until the preview result is satisfying. Click OK to
apply the selected values to the whole dataset. The volume values in the
spinboxes are in pixels, the numbers marked Physical are in cubic micrometers.
In the lower right corner of the dialog window you can find information about
the number of removed/remaining objects.
Remark 4 Note that removing all boundary objects using the Object Filter
distorts subsequent statistics on geometric characteristics of the remaining
objects as the remaining non-border objects form a size-biased sample. This is
due to the fact that large and long objects are more likely to be cut by the
image border and thus to be discarded. Statistics on geometric characteristics
should rather be based on the Object Features measured on the unfiltered
label image together with a treatment for edge effects, see Section 8.5.8.
8.4.13 Map Subfield Feature To Image
This function is selectable for images for which the Subfield Features have
previously been computed. Choose the feature of interest from the dropdown
list and click OK. The result is a new GRAYF image containing the numerical
value of the selected feature as pixel gray value in each subfield (square for 2D
data, cube for 3D data).
The option Crop result in the dialog (see Figure 38) allows to adjust the size of
the resulting image such that it does not contain pixels for which no feature
data is available. This will happen whenever the image’s side lengths are not an
integral multiple of the subfields’ side lengths. Usually, you should leave this
option enabled.
8.4.14 Map Subfield Fiber Directions To Image
This function will be selectable for images for which you have previously
computed the (Sub-)Field Fiber Directions. The measurement to be visualized
can be chosen via a dialog in a similar manner as described for Map Subfield
Feature To Image.
MAVI user manual
75
Map Subfield Fiber Directions To
Image
Figure 37: Object Filter dialog window.
MAVI user manual
76
Map Subfield Fiber Directions To
Image
Figure 38: Map Subfield Feature To Image dialog to select a feature whose
value will be written to a new image.
MAVI user manual
77
Analysis
8.5
Analysis
Quantitative geometric analysis of components or objects in images forms the
core of MAVI. This submenu offers a wide variety of analysis functions, starting
with basic gray value Image Statistics and ending with Integrated Plugins
offering user friendly short cuts for complex image processing chains for
Particle Separation and Cell Reconstruction.
Based on efficient algorithms for determining the intrinsic volumes or their
densities, the Field Features, Subfield Features, and Object Features yield
geometric characteristics for components of microstructures or objects like
particles or pores or fibers.
For special classes of structures the Open Foam Features, (Sub-)Field Fiber
Directions, and Geometric Tortuosity yield mean characteristics of open
foams, local fiber orientations, and a means to quantify how "warped" the pore
space is.
The Subfield Features as well as Area Fraction Profile and Grayvalue
Profile allow to detect and quantify inhomogeneities.
Moreover, spectral analysis functions like Covariance and Bartlett spectrum
are available.
8.5.1
Image Statistics
Image Statistics shows the gray value histogram of the image as well as
minimum, maximum, and mean gray value and the standard deviation. The
histogram can be exported in CSV format by clicking on the Export button.
8.5.2
Rotation Mean
The rotation mean of an image is a one-dimensional function giving as ith value
the average value of all pixels at the following distance from the image center:
ismin d,
where smin = min{s1 , s2 , s3 }
(5)
denotes the minimum of the lattice spacings in the three coordinate directions.
The user defined parameter d controls how fine the rotation mean is discretized.
The default value d = 1 results in a bin width of one pixel (for an isotropic
lattice) while d = 0.25 samples the distances four times finer. Note however that
for distances smaller than 2smin this fine sampling is impossible due to lack of
values.
MAVI user manual
78
Field Features
8.5.3
Field Features
Field features measure geometric characteristics of the foreground of MONO
images. It is not neccessary that the foreground consists of isolated and
topologically simple objects. It is however assumed that the foreground pixels
form a discrete representation of a subsample of a macroscopically
homogeneous microstructure. That is, the structure extends beyond the image
without changing its average properties.
The field features are essentially the intrinsic volumes and their densities, also
known as Minkowski functionals or quermass integrals. For a mathematical
definition of the field features and their interpretation for constituents of
macroscopically homogeneous microstructures see [12, Chapter 5]. The
algorithm for measuring the field features goes back to [6] and is
comprehensively described in [10]. Note that all measurements rely on local
2x2x2 pixel configurations only and do not require a surface meshing. Edge
effects therefore occur at the last slices of the image only. These border
configurations are treated differently such that the final results are free of edge
effects.
Generalized projections onto one- and two-dimensional subspaces are
computed as intermediate steps of the measurement of the integral of mean
curvature and the surface area, respectively. Here ’generalized’ means measured
with multiplicity as illustrated in Figure 39 below. These generalized projections
Figure 39: Generalized projection of a bent curve: The length of the projection
of this curve onto the x-axis is a+b+c. The length of the generalized projection is
however a+3b+c since all three ’layers’ contribute.
carry information about anisotropies and preferred directions. Discrete
distributions of the fiber direction in the typical fiber point and of the surface
normal direction in the typical surface point can be deduced. They are
concentrated on the respective set of 13 discrete directions given by the lattice
unit cell (see 4.4, Tables 2 and 3) and represented in the output of the field
MAVI user manual
79
Field Features
features by the weights for these directions.
For special cases of microstructures the field features have to be interpreted as
appropriate, the Open Foam Features are provided as special function.
The base unit for all features is one meter, the only exception to this rule is an
image without pixel spacings (see 4.1.1). In this case, the base unit is one pixel.
The following Table 11 lists the Field Features measured for 3D images
together with their description and the unit they are measured in.
Name as in attributes view
total volume V
volume of foreground
volume density
surface
surface density
surface fractal dimension
area of projections
orientation distribution of surface normals
integral of mean curvature
integral of mean curvature density
integral of mean curvature fractal dimension
lengths of projections
unit
[m3 ]
[m3 ]
[
]
2
[m ]
[m−1 ]
[
]
[m2 ]
[
]
description
of the image (foreground and background)
[m ]
[m−2 ]
[
]
integral of Germain’s curvature M
MV = M/V
box dimension related to M
[m
]
on lines in the 13 normal directions (Table 3)
orientation distribution of fibers (edges)
specific fiber length
fiber aspect ratio
integral of total curvature
total curvature density
Euler number for all neighborhoods
[
]
[m−2 ]
[
]
[
]
[m−3 ]
[
]
Euler density
mean chord lengths
[m−3 ]
[m ]
mean chord lengths mean
structure model index
[m
[
relative frequencies
(6)
(7)
integral of Gaussian curvature
integral of total curvature /V
Euler numbers w.r.t. 26, 14.1, 14.2, and 6
adjancency systems
Euler number /V
mean chord lengths in 13 directions (Table 2), (8)
mean of the 13 mean chord lengths
shape factor, (9)
]
]
volume fraction VV , foreground volume /V
area of the inner surface S
specific surface area SV = S/V
box dimension related to S
perpendicular to the 13 directions (Table 2)
relative frequencies
Table 11: Field Features measured for 3D images.
The density of the integral of mean curvature MV yields the specific fiber length
LV :
LV = 4MV /π
(6)
MAVI user manual
80
Field Features
Comparison of the generalized projection lengths in the coordinate directions
can reveal dilation or compression of a fibrous structure: The ratio
2projection lengthz /(projection lengthx + projection lengthy )
(7)
equals 1 for an isotropic structure, variations indicate dilation or compression in
z-direction. Of course, these characteristics are meaningful for fiber structures
only.
The mean chord lengths (MIL) in the 13 discrete directions are deduced from the
areas of projection via
(8)
MIL = VV /projection area.
The structure model index SMI – a shape factor for components – can be
derived from volume fraction, specific surface area and density of the integral of
mean curvature [11]
SMI = 12VV MV /S2V .
(9)
It assumes values 0 for ideal planar structures, 3 for spherical cylinders and 4 for
non-overlapping balls.
The following Table 8.5.3 lists the Field Features measured for 2D images
together with their description and the unit they are measured in.
Name as in attributes view
total area A
area of foreground
area density
boundary length
boundary length density
unit
[m2 ]
[m2 ]
[
]
[m ]
[m−1 ]
lengths of projections
integral of total curvature
total curvature density
Euler number
Euler density
mean chord lengths
mean chord lengths mean
[m ]
[
]
[m−2 ]
[
]
−2
[m ]
[m ]
[m ]
description
of the image (foreground and background)
area fraction AA , foreground area /A
specific boundary length, total length per
area
in the 8 discrete directions
integral of Gaussian curvature
integral of total curvature /V
Euler number w.r.t. 6.1 adjancency system
Euler number /A
mean chord lengths in 8 discrete directions
mean of the 8 mean chord lengths
Table 12: Field Features measured for 2D images.
Remark 5 Both field features and object features are based on the intrinsic
volumes. Nevertheless, the total values of e. g. volume or surface area in the
field features might differ from those obtained by the object features. This is
MAVI user manual
81
Subfield Features
due to the fact that the latter are measured just for the present object, while
MAVI implies the whole space to be filled with an on average similar structure
when measuring the Field Features of the foreground.
Field features and object features coincide if you ensure that the foreground
does not intersect the image border, e. g. by Manipulation:Padding.
Precision of the Field Features’ measurement There is a variety of sources
for measurement errors when determining the Field Features, starting with the
imaging conditions, continuing with discretization effects, and finally
binarization. It is therefore impossible to quantify the overall measurement error.
However, there are some rules of thumb:
Specific surface area as well as the density of the integral of mean curvature are
accompanied by the respective so called fractal dimensions. These values
indicate resolution dependence of the measurements. That means, if the
surface fractal dimension is larger than 2.2, then the resolution of the image
was not sufficient to measure the specific surface area properly. The same holds
true for values of the integral of mean curvature fractal dimension deviating
significantly from 1.
Large differences of the Euler numbers with respect to the different discrete
neighborhoods are another indicator for the resolution being not sufficient for
the Euler number measurement on the given structure.
If you are nevertheless sure that the resolution was good enough then apply a
smoothing operation before or after segmentation.
Robustness of the field features decreases with their dimension. The values for
volume fraction or specific surface area can still be trusted even if the integral of
mean curvature fractal dimension is 2 or the Euler numbers differ strongly. On
the other hand, if the surface fractal dimension is 2.5, then not only the
measured specific surface area but also the integral of mean curvature and the
Euler number should be regarded with suspicion.
8.5.4
Subfield Features
The Subfield Features calculate the Field Features locally, thus allowing to
quantify structural differences within one sample without cutting up the
respective image. As the Field Features, the Subfield Features are measured on
the foreground of MONO images. The square (2D) or cubic (3D) subfields of
user-specified size form a tiling of the image. That is, subfields do not overlap
and pixels at the image edges not being covered are not taken into account.
MAVI user manual
82
Area Fraction Profile
The side length of the subfields needs to be specified by the user in units of
meters, see Figure 40.
Figure 40: Subfield Features: dialog for specifying the subfield size.
For a list of the resulting measurements in each subfield see the Field Features.
In the resulting view, each subfield can be found by the coordinate of its first
point, in [pixel]. Subfield Feature results can be saved using Field Features as
CSV from the File:Export menu and visualized using Map Subfield Feature
To Image from the Transformation menu.
8.5.5
Area Fraction Profile
For macroscopically homogeneous microstructures, the volume fraction
coincides with the area fraction in a 2D slice. Conversely, changes of the area
fraction indicate inhomogeneities, reveal layered structures or structures with a
gradient.
The Area Fraction Profile calculates for MONO images the area fraction of
each slice in the coordinate directions. The results can be exported as a CSV file
via File:Export: Area Fraction Profile as CSV.
Optionally, a mask image can be provided. In that case, the area fraction is
obtained as area of foreground within mask divided by area of mask in the
current slice. This allows e. g. to quantify changes of material density within a
complex shaped part.
For 2D images, the Area Fraction Profile yields the profile of line fractions.
That is, the length fraction of section lines covered by foreground.
MAVI user manual
83
Grayvalue Profile
8.5.6
Grayvalue Profile
The Grayvalue Profile calculates for gray value images the mean gray value in
each slice in the coordinate directions. Strong or systematic changes indicate
global gray value fluctuations either due to imaging artefacts like beam
hardening or due to structural inhomogeneities like layers or a gradient.
The Grayvalue Profile can be exported as a CSV file via File:Export:
Grayvalue Profile as CSV.
Optionally, a mask image can be provided. In that case, only the gray values of
those pixels are averaged that are foreground in the mask image.
For 2D images, the Grayvalue Profile yields the mean gray value in each line in
the coordinate directions.
8.5.7
Granulometry
This function performs a fast exact spherical granulometry transform on MONO
images. The user defines whether foreground (default) or background are
analyzed.
The spherical granulometry assigns each pixel the diameter of the largest ball
completely contained in the foreground or background, respectively, and
covering the pixel. It thus yields the local structure thickness or size or the local
pore size.
More precisely, let X be the structure to be analyzed, Br the ball of radius r,
and X ◦ Br the morphological Opening of X with Br . The granulometry
distribution function
G(d) = 1 −
1 − VV (X ◦ Bd/2 )
,
1 − VV (X)
d≥0
yields a volume weighted generalized size distribution. It can be measured from
the MONO image representing X as foreground by successive morphological
openings with growing balls until all foreground pixels have disappeared. Finally
each pixel x is assigned the diameter of the ball used as structuring element in
the step where x disappeared for the first time. The resulting image is called
ultimately opened or granulometry image. For the actual calculation a more
efficient algorithm based on the medial axis coinciding with the crest lines of
the Euclidean distance image is used.
Results are the granulometric curve (also called pattern spectrum ) or a GRAYF
image holding the local sizes (diameters). The granulometric curve is the
MAVI user manual
84
Object Features
histogram of the volume weighted size distribution. Peaks in this curve indicate
the predominant structure or pore size. Results can be saved using File:Export:
Granulometric Data as CSV.
Remark 6 This algorithm’s run time depends highly on the local structure
thickness or pore size and can increase to an impracticable length. Consider
alternatives like the Euclidean Distance Transformation if diameters larger
than 200 pixels are expected.
8.5.8
Object Features
An object in a 3D image is a connected component representing for instance a
particle, a pore, a cell, or a fiber. Object Features can be applied to MONO
images or label images as obtained by a Labeling or the Watershed
Transformation. In the first case, the foreground is treated as the one and only
object. In the case of a label image, the geometric characteristics of each of the
objects are computed.
The volume is measured by simply counting the pixels of the object. For details
on the measurement method for the intrinsic volumes refer to the Field
Features section. The following Table 8.5.8 lists the features measured for
objects in 3D images together with their description and the unit they are
measured in. The bounding box is the smallest cuboid with axes parallel to the
coordinate axes and containing the object.
For an object with volume V the diameter d of the equal volume ball is just
p
d = 3 6V /π.
(10)
Note that the diameters of an object in the 13 normal directions (Table 3)
coincide with those for its convex hull formed by planes perpendicular to those
directions.
The Euler number is a topological characteristic. The following examples
illustrate its meaning: The Euler number of an object having no holes (‘caves’)
and ‘tunnels’ is 1, if the object has a hole then its Euler number is 2, if the
object is topologically equivalent to the torus, we get 0, if the object has two
‘caves’, we get Euler number 2, and if it has two ‘tunnels’ one obtains -1. Euler
number and the integral of total curvature carry the same information about the
object. They differ by the constant factor 4π.
Additional characteristics for 3D objects like length, width, and thickness or
elongation are contained in the module Particle Features.
For 2D objects, MAVI determines the following characteristics:
MAVI user manual
85
Object Features
Name as in attributes view
center
diameter of corresponding ball
diameters
unit
[Pixel]
[m ]
[m ]
Euler number
integral of mean curvature
integral of total curvature
no. of voxels
object
offset
on border
shape factor 1
shape factor 2
shape factor 3
size
status
[
]
[m ]
[
]
[
]
[
]
[Pixel]
true/false
[
]
[
]
[
]
[Pixel]
surface area
surface of convex hull
[m2 ]
[m2 ]
volume
volume of convex hull
[m3 ]
[m3 ]
weight Miles-Lantuejoul
[
]
description
central pixel of bounding box
diameter of equal volume ball, (10)
lengths of projections on lines in the 13 normal
directions (Table 3)
w.r.t. 14.1 adjancency system
integral of Germain’s curvature M
integral of Gaussian curvature
number of the object, 0 refers to background
lower left front corner of bounding box
indicates whether object touches image border
sphericity, (11)
(12)
(13)
edge lengths of bounding box
if not OK, a numerical problem occured, e. g. due
to the object being too small
area of inner and outer surface
surface area of convex hull formed by planes perpendicular to the 13 normal directions (Table 3)
volume of convex hull formed by planes perpendicular to the 13 normal directions (Table 3)
edge correction factor, (14)
Table 13: Object features measured for 3D images.
Remark 7 When measuring the Object Features, MAVI minimizes the mean
discretization error for an object in “general” position. Therefore, the quality of
the measurements can not be judged by comparing to theoretical values for
objects in a particular position. In particular, the surface area and the integral of
mean curvature of a cube parallel to the coordinate axes are underestimated,
while they are overestimated for the same cube rotated e. g. by 20◦ about the
z-axis, flipped, and rotated by 20◦ about the z-axis again.
Isoperimetric shape factors The isoperimetric shape factors
√
√
f1 = 6 πV / S 3
f2 = 48π 2 V /M 3
f3 = 4πS/M 2
(11)
(12)
(13)
are derived from the intrinsic volumes, see Field Features. These shape factors
are normalized such that f1 = f2 = f3 = 1 for a ball. Deviations from 1 thus
MAVI user manual
86
Object Features
Name as in attributes view
area
boundary length
center
diameter of corresponding circle
diameters
Euler number
integral of curvature
mean chord length
no. of pixels
object
offset
on border
size
status
unit
[m2 ]
[m ]
[Pixel]
[m ]
description
[m ]
[
]
[
]
[m ]
[
]
[
]
[Pixel]
true/false
[Pixel]
lengths of projections on lines in 8 directions
w.r.t. 6.1 adjancency system
integral of Gaussian curvature
central pixel of bounding box
diameter of equal area circle, (10)
number of the object, 0 refers to background
lower left front corner of bounding box
indicates whether object touches image border
edge lengths of bounding box
if not OK, a numerical problem occured, e. g. due
to the object being too small
Table 14: Object features measured for 2D images.
describe various aspects of deviations from spherical shape. The shape factor f1
is often called sphericity. Examples for some simple objects and the theoretical
values of the shape factors 1-3 for them are given in Table 15. More shape
information can be derived e. g. from the ratio of the volumes of an object and
its convex hull or the ratio of minimal and maximal diameter.
Object
ball
prolate spheroid
oblate spheroid
cylinder
cylinder
cube
cuboid
description
long:short axis as 2:1
long:short axis as 2:1
height:diameter as 1:2
height:diameter as 1:1
length:depth:height as 1:2:3
f1
1.00
0.90
0.87
0.75
0.82
0.72
0.62
f2
1.00
0.76
0.80
0.68
0.71
0.57
0.42
f3
1.00
0.89
0.94
0.93
0.91
0.85
0.78
Table 15: Isoperimetric shape factors. Examples.
Edge correction When determining distributions of characteristics like
volume or diameter of objects in an image, objects intersecting the boundary
should not be included. However, bigger objects have a higher probability of
intersecting the boundary than smaller ones. The Miles-Lantuejoul weight
corrects boundary effects caused by this size dependent sampling. Each object is
assigned a weight proportional to the reciprocal of its probability of being
MAVI user manual
87
Open Foam Features
sampled. Roughly speaking, large objects not hitting the boundary are hard to
observe. Once this unlikely event happens, a high weight is assigned to this
object:
total volume/volume of window eroded by bounding box of object.
(14)
This factor appears as “weight Miles-Lantuejoul” in the Object Features. For
border intersecting objects, it is 0. When estimating moments etc. from the
object features, sum all values multiplied by the weight. Subsequently,
normalize by dividing by the sum of the weights, see [12, Section 5.2.6].
8.5.9
Open Foam Features
The Open Foam Features are mean characteristics of open foams, calculated
from the Field Features using a model assumption: The typical cell of the open
foam is assumed to be up to a scaling factor the same as the typical cell of one
of the 5 choosable models. The Laguerre tessellation models are generated by
(a) Poisson Voronoi
(b) Hardcore Voronoi
(c) Laguerre, cv 0.2
(d) Laguerre, cv 2.0
(e) Weaire-Phelan foam
Figure 41: Models used by the Open Foam Features. (a)-(d) – random tessellation models.
force biased packings of balls. The ball volumes follow a log normal distribution,
with low (0.2) and high (2.0) coefficient of variation cv, respectively. See [18]
and [12, Section 7.6] for theoretical background.
Open Foam Features can be measured from 3D MONO images only. The
foreground should represent the strut system of an open foam. That is, the
structure is supposed to have no closed walls and the struts have to be closed.
Hollow struts or isolated closed walls should therefore be closed or removed,
respectively, using morphological transformations 8.4.2.
Remark 8 Open Foam Features do NOT support 2d images.
The scaling factor can be deduced from the strut length density (total strut
length per unit volume) or the density of the Euler number, see [17] for details.
In a simulation study, the Euler number based scaling proved to be more precise
than the strut length based one. It is therefore now the default scaling used by
MAVI user manual
88
Geometric Tortuosity
MAVI. The strut length based scaling is included for backward compatibility
only.
System
Cells
Faces
Edges
Nodes
characteristic
mean cell volume
mean number of cells per unit volume
mean number of cells per unit length
ppi-value (pores per inch)
mean cell surface area
mean cell diameter
mean cell chord legth
mean number of faces per unit volume
mean face diameter
mean face area
mean number of edges per unit volume
mean edge diameter
mean edge circumference
mean edge cross section area
length density of the edges
mean number of nodes per unit volume
unit
[m3 ]
[m−3 ]
[m−1 ]
[
]
2
[m ]
[m ]
[m ]
[m−3 ]
[m ]
[m2 ]
[m−3 ]
[m ]
[m ]
[m2 ]
[m−2 ]
[m−3 ]
Table 16: Characteristics calculated by the Open Foam Features.
The Open Foam Features compute mean values of the cells, faces, and edges of
the foam structure in the image. In order to analyze properties of individual cells
or to estimate empirical distributions, the cells have to be reconstructed by an
image analytical Cell Reconstruction. Subsequently, the Object Features can
be calculated on the cell system. The cell reconstruction process is applied to the
open nickel foam and zinc foam examples.
8.5.10 Geometric Tortuosity
The geometric tortuosity of a porous medium is the ratio of the length of the
shortest path between two points within the foreground component to their
Euclidean distance.
All algorithms measuring the geometric tortuosity are based on the concept of
geodesic distance of two foreground pixels [19]. It’s the length of the shortest
digital path connecting the two pixels and not leaving the foreground
component. MAVI measures on MONO images, for each coordinate direction,
the ratio of geodesic and Euclidean distance of border foreground pixels to the
opposite image border. The geometric tortuosity is then calculated by a
weighted averaging of all such ratios. More precisely, the ratios are squared and
weighted by the frequency of their occurence.
MAVI user manual
89
(Sub-)Field Fiber Directions
The results can be exported as a CSV file via File:Export: Geometric
Tortuosity as CSV.
8.5.11 (Sub-)Field Fiber Directions
The (Sub-)Field Fiber Directions estimate the distribution of the local direction in
the typical point of a fiber system. Contrary to the majority of the Analysis
functions they operate on GREY8 images. Images to be analyzed should contain
a randomly distributed fiber system and the fibers should have a larger gray
value than the matrix or pore space pixels. Furthermore, the fibers should have
an approximately constant and known radius. These assumption are in
particular fulfilled by µCT images of glass fiber reinforced composites. A typical
example is discussed in 7.3.
Remark 9 (Sub-)Field Fiber Directions do NOT support 2D images.
The (Sub-)Field Fiber Directions can operate on the whole image as the Field
Features or on subvolumes of user defined size as the Subfield Features.
When calling the (Sub-)Field Fiber Directions, the user has to specify the radius
of the fibers in meters, see the dialog in Figure 42. The fiber diameter should be
resolved with 5-10 pixels. The function proved however to work at the absolute
minimum of 3 pixels per fiber diameter too, see the tutorial 7.3. The (Sub-)Field
Fiber Directions calculate the local fiber direction in each pixel without
segmentation of individual fibres. MAVI uses the method based on the Hessian
matrix of second order partial derivatives of the gray values as described in [15].
The gray value image is smoothed by a Gauss Filter with parameter adjusted to
the fiber radius. Subsequently, the Hessian matrix H(x) of second order gray
value derivatives of the smoothed image is calculated in each pixel x. The
eigenvectors of H carry information about directions of the fiber system at x.
The least gray value variation is expected along the fibre. Thus the eigenvector
corresponding to the smallest eigenvalue of the Hessian matrix H(x) at x is
interpreted as the local direction ν1 , ν2 , ν3 in pixel x.
Finally, the result is restricted to the pixels belonging to the fiber system using
either a mask image provided by the user or an automatically computed mask
image. The former is specified by the user via the dialog (Figure 42) while the
latter is obtained by global thresholding (see 8.3.1). The threshold according to
Otsu is multiplied by 1.25 (default) or a user defined value.
Remark 10 The rational behind the increased threshold is the observation that
the resulting directional distribution is not heavily distorted if the cores of the
fibers are regarded only.
MAVI user manual
90
(Sub-)Field Fiber Directions
Figure 42: (Sub-)Field Fiber Directions: dialog for entering the fiber radius and
an optional mask image.
The averaged outer product of the components ν1 , ν2 , ν3 is an estimator for the
so-called second order orientation tensor widely used in simulation of materials
properties in order to incorporate anisotropic local behaviour:
X
aij =
νi (x)νj (x).
(15)
x∈image and fiber system
The eigenvector corresponding to the largest eigenvalue l3 of the orientation
tensor now yields the principal fiber direction. This principal direction is however
only meaningful if the fiber system features a preferred direction at least locally
(in the subvolumes). If the fibers are isotropically distributed, then the three
eigenvalues coincide. The more anisotropic the fiber system gets, the stronger
differ the maximal and the minimal eigenvalue, l3 and l1 , respectively. Thus, a
measure for the degree of anisotropy is
1 − l1 /l3 .
(16)
This expression assumes values between 0 and 1, where 0 corresponds to an
ideal isotropic fiber distribution, 1 to either all fibers being parallel to the main
fiber direction or all fibers being parallel to a plane.
MAVI user manual
91
(Sub-)Field Fiber Directions
The (Sub-)Field Fiber Directions yield the following information for the whole
field and – if applicable – for each subfield: If subfields have been specified, the
Name
Anisotropy value
Eigenvalues of the orientation tensor
Orientation tensor
Mean fiber direction
Status
description
(16)
l3 ≥ l2 ≥ l1
3x3 symmetric matrix (15)
eigenvector to l3 , princial fiber direction
Warns about numerical problems, e. g. due to
the subfield not containing any pixels belonging
to the fiber system. Should state "OK".
Table 17: (Sub-)Field Fiber Directions results.
Averaged Profile holds additionally the characteristics from Table 8.5.11
averaged over all subfields in the same layer. That is, for each coordinate
direction, the characteristics are averaged over all subfields with the same
respective coordinate.
If subfields have been specified, MAVI offers the possibility to visualize selected
fiber measurements using Map Subfield Fiber Directions To Image from the
Transformation menu. Results can be saved using File:Export: Subfield Fiber
Directions as CSV.
MAVI user manual
92
Spectral Methods
8.5.12 Spectral Methods
Second order characteristics describe fluctuation of microstructures. They as
well as their counterparts in frequency space can be measured from 3D images
of the microstructures. The fast Fourier transform allows to determine these
quantities like the covariance function fast and efficiently.
Convolution The convolution is a local moving average operation involving
two images, where the second image serves as a filter mask. For each pixel in
the first image, the second image is shifted so that its origin is located at the
given pixel and the average is computed over all pixels covered by the second
image. For calculation of the average, pixel values in the first image are
weighted by the corresponding values in the second image.
The convolution of two images is equal to the Inverse Fast Fourier
Transformation of the product of their Fast Fourier Transformations.
Accordingly, the implementation of the convolution in MAVI is based on Fourier
methods using the FFTW 3.0.1 library [3].
Both images have to have the same dimensions.
Covariance The covariance of two images (a and b) is basically the Inverse
Fast Fourier Transformation of the product of the Fast Fourier
Transformation of image a and the Complex Conjugate of the Fast Fourier
Transformation of image b. High gray values in the resulting image indicate
strong similarity in the structures of the input images.
Covariance (Rotation Mean) Rotation mean of the Covariance – a
one-dimensional function giving as ith value the average covariance at distance
ismin d where smin is the minmum of the lattice spacings in the coordinate
directions. Shortcut for Rotation Mean applied to the Covariance saving
computation time and memory. The parameter d is determined by the user with
values 1 (default) or 0.25. For details see Section 8.5.2. The result can be
exported using File:Export: Rotation Mean of Covariance as CSV.
Autocovariance The autocovariance of an image is the Covariance of this
image with itself, that is the Inverse Fast Fourier Transformation of its
Power Spectrum.
MAVI user manual
93
Spectral Methods
The autocovariance of a MONO image yields an unbiased estimator for the
covariance of the foreground seen as realization of a spatially homogeneous
random closed set (see [5]). It is called two-point probability function or
two-point correlation function, too.
Autocovariance (Rotation Mean) Rotation mean of the Autocovariance –
a one-dimensional function giving as ith value the average autocovariance at
distance ismin d where smin is the minmum of the lattice spacings in the
coordinate directions. Shortcut for Rotation Mean applied to the
Autocovariance saving computation time and memory. The parameter d is
determined by the user with values 1 (default) or 0.25. For details see
Section 8.5.2. The result can be exported using File:Export: Rotation Mean
of Autocovariance as CSV.
Bartlett Spectrum The Bartlett spectrum of a MONO image is basically the
Fast Fourier Transformation of the Autocovariance of that image. It is the
counterpart to the covariance in frequency space and – contrary to the Power
Spectrum – not distorted by the image size. It solely captures fluctuations of
the analyzed microstructure.
Bartlett Spectrum (Rotation Mean) Rotation mean of the Bartlett
Spectrum – a one-dimensional function giving as ith value the average Bartlett
Spectrum at distance ismin d where smin is the minmum of the lattice spacings in
the coordinate directions. Shortcut for Rotation Mean applied to the Bartlett
Spectrum saving computation time and memory. The parameter d is
determined by the user with values 1 (default) or 0.25. For details see
Section 8.5.2. The result can be exported using File:Export:Rotation Mean of
Bartlett Spectrum as CSV.
Bartlett Spectrum From Autocovariance The Bartlett Spectrum can be
calculated directly from the Autocovariance. If the latter has already been
computed Bartlett Spectrum From Autocovariance saves time and memory.
MAVI user manual
94
View
8.6
View
8.6.1
Slice View
The slice view is the standard view in MAVI. For 2D images, it just shows the
image, the coordinate axes and a scale. For 3D images, it provides three view
windows showing 2D slices orthogonal to the coordinate axes and a fourth
window holding a cuboid illustrating the location of the slices within the 3D
image.
Exactly one of the three view windows is active at a time. Activation is indicated
by a yellow star in the lower left corner of the view window, see Figure 43.
Figure 43: MAVI’s Slice View for a 3D image.
MAVI supports realistic representation of anisotropic grids, that is, if an image
has different spacings along the coordinate axes, this will be reflected in the
representation. Figure 44 illustrates this using the example from Figure 43 with
different spacings.
MAVI user manual
95
Slice View
Figure 44: MAVI’s Slice View representation of a 3D image with anisotropic
spacings.
Slice View Control Toolbar The toolbar
offers functionality to
control or modify the view windows.
You can zoom in to or out of the currently activated slice view by clicking on
one of the
buttons. The button resets the current view to a zoom level of
100% while the button zooms to the maximum keeping the current slice
completely visible.
For 2D images, these are the only buttons present in the toolbar.
For 3D images, the first five buttons control what the slice viewer actually
shows. Clicking one of the first four buttons to the left maximizes the
corresponding view window. Clicking on the fifth button will restore the
standard slice view layout with the four windows.
The
buttons facilitate navigation through the image.
The index of the slice visible in the currently activated slice view is displayed. You
MAVI user manual
96
Slice View
can scroll through the stack of slices either by moving the slider underneath the
display or by clicking the up or down buttons. The buttons with one arrow
move on by one slice and the buttons with two arrows move by ten slices.
The
button offers the chance to disconnect the slice-slider from the update
of the slice. It is relevant when working remotely, only. By default, it is set to
“connect”. If it is turned to “disconnect” , the slider can be moved to the
desired slice without updating the view. The view is not updated until the slider
is disengaged.
Slice View Control Window The view control window contains several tabs
containing further functions to control and modify the view windows. A tab can
be activated simply by clicking on it.
The Info tab contains information on the current view window, such as location
of slices and mouse position within the image, see Figure 45(a).
The Settings tab (Figure 45(b)) contains the functions
Reset window: undoes any translations, magnifications or changes of
background color on the currently activated view window
Set BG Color: offers a choice of background colors via a color selection dialog
Save current slice view: saves a 2D image of the current slice view in the
active view window including the view filter if applicable but not the zoom.
Available file types are BMP, PNG and XPM. An additional dialog offers the
possibility to decide whether a scale should be drawn into the 2d image.
The Slice Show tab provides automatic scrolling through the slices of a 3D
image in the current view window, see Figure 45(c). Adjust the scrolling speed
in the Delay field. The animation will start in the currently activated view
window and can be shifted to another one by simply clicking into the desired
window. Using the Create Movie button, an animated walk through the image
in the direction given by the active view window can be saved. On Windows
operating systems, an avi file is created. On Linux operating systems, a dialog
pops up allowing to save the slices as bmp or png files to a designated directory.
A Windows compatible avi can then be created using the following expression
on the command line:
mencoder "mf://*.bmp" -mf fps=12 -o output.avi -ovc lavc -lavcopts
vcodec=wmv1
The View Filter tab provides filters to modify the color representation of an
image in the Slice View, Figure 45(d). Applying a view filter does not change the
actual pixel values of an image but simply offers another way of looking at the
data. See Figures 46 and 47 for examples of view filters and their effect. As can
be seen in Figure 47, gray values may be difficult to distinguish visibly. In such a
MAVI user manual
97
Image Attributes View
case, a false color representation adds clarity and information can be extracted
more easily from the image on a visual basis. Table 18 lists the color tables
available for viewing gray scale images. For COMPLEX images, the view filter
tab provides the choice between real and imaginery part. For images of type
GRAYF (float) the pixel values are mapped to an integer gray value range that
can be displayed.
(a) Info tab
(b) Settings tab
(c) Slice Show tab
(d) View Filter tab
Figure 45: The tabs in the View Control Window of the Slice View.
8.6.2
Image Attributes View
In addition to the actual pixel gray values, an image in MAVI contains various
attributes, which can be viewed by clicking on Image Attributes View.
Moreover, Analysis results like Field Features or Object Features will be
visible here. It is however not possible to copy data from the Image Attributes
View window. Image attributes and features can instead be exported in CSV
format via File:Export.
8.6.3
Volume Rendering View
Volume Rendering View allows visualization of MONO and GRAY8 3D images.
The Volume Rendering View window opens in the central workspace when the
function is chosen via the View menu. The corresponding list entry consisting
of the view icon, the view label, and an additional description appears in the
view navigator window.
MAVI user manual
98
Volume Rendering View
(a) MONO image, default
(b) Tomographic view
(c) MONO image, materialographic
(d) Materialographic view
Figure 46: View filters for MONO images in the Slice View.
Volume Rendering View step-by-step
(1) Start the Volume Rendering View, the corresponding window opens,
Figure 48(a)
(2) Check the position of your sample w.r.t. the view
(3) Change the background color to check whether imaging artefacts or
noise might disturb the final rendering, Figure 48(b)
(4) Adapt the transfer function to the gray value distribution in your image:
Adjust the red, green, blue, and alpha channels, Figure 48(b)
(5) Save the thus derived color table in between.
(6) Use reset to return to the initial settings if needed.
(7) Adapt the blend scale, Figure 48(c)
(8) Adjust the lights, Figure 48(d), with the help of the light source positions
if needed, Figure 48(e).
Volume Rendering View toolbar The Volume Rendering View window has
its own menu. Its icons allow to move, rotate, zoom the visualized image. The
view control window now contains now tools and functions relevant for the
Volume Rendering View.
On the top-left side of the widget, the
button starts animation of the image,
rotating about the X axis. Another click on that button stops the animation.
Moving the scrollwheel on top of the widget zooms in and out. On the right
hand side of the zoomwheel, there is a group of three buttons which select the
action to be performed when the user moves the mouse over the render area
with a mouse button pressed: The
button rotates the image around
different axes, depending on the mouse button that is pressed. The different
MAVI user manual
99
Volume Rendering View
(a) GRAY8 image
(e) GRAY8 image
(b) Gray scale view
(c) GRAY8 image, 12bit mode
(d) 12bit view
(f) GRAY8 image, blue-to-red mode
Figure 47: View filters for GRAY8 images in the Slice View.
axes with their corresponding buttons are listed in the following Table 19. The
button moves the image inside the render area. The
button zooms in
and out of the image when the mouse is moved in vertical direction with the
left mouse button pressed. This is the same operation as possible using the
upper scrollwheel. On the bottom of the render widget, there are three
scrollwheels which rotate the displayed image about the three different axes.
The first one rotates the image view about the X axis, the second one about the
Y axis and the third about the Z axis.
Volume Rendering View control window On the lower left hand side of
MAVI’s main window there is a toolbox, containing different advanced tools to
modify the visual appearance of the rendered image.
Transfer function The first and most important entry in the toolbox is the
transfer function editor controling how the gray values of your image are
MAVI user manual
100
Volume Rendering View
Color table
Description
Gray Scale
resets the display to gray scale (default)
12 Bit
Scale
Gray
shows only the first 12 bits of the pixel gray values,
for GRAY16 images where not the full 16bit gray
value range is used (like in some images from computed tomography), has no effect on GRAY8 images
12 Colors (for
label images)
for a better display of label images, assigns one of
12 colors to each label, adjoining labels receive distinct colors
125 Colors (for
label images)
for a better display of label images, assigns one of
12 colors to each label, adjoining labels receive distinct colors
Power Spectrum
(blue - red)
power spectrum false color palette from blue to red
Power Spectrum
(red - blue)
power spectrum false color palette from blue to red
Rainbow
rainbow false color palette
Color range
Red Scale
Green Scale
Blue Scale
Table 18: Color tables in the Slice View. Color tables are applicable to gray
scale integer valued image only, that is to images of type GRAY8, GRAY16, or
GRAY32.
mapped onto the displayed colors red, green, and blue (RGB), and the alpha (A)
channel. The RGB channels are separate functions mapping a gray value to the
corresponding color channel. These three fundamental colors are mixed to
generate the pixel’s color. The alpha channel governs the transparency of the
pixel, that is, how much of the pixel behind that pixel is visible.
Initially, the transfer function is the identity function resulting in a gray level
visualization with decreasing transparency. Black source pixels are mapped to be
fully transparent and white source pixels to appear totally opaque
(non-transparent).
In the background of the function widget, the gray value histogram of the
image is shown.
To modify the transfer function, first select one of the channels red, green, blue,
and alpha below the function widget. Clicking on the function graph creates a
MAVI user manual
101
Volume Rendering View
Button
Left button
Right button
Mouse movement
vertical
horizontal
vertical
horizontal
Rotation axis
Y axis
X axis
Z axis
X axis
Table 19: Axis and button relation in rotation mode.
new control point, which can be moved to modify the run of the curve. The
rendered image is updated each time the mouse button is released. In case the
update fails, then change the blend mode shortly and return to your previous
blend mode. Existing control points for the selected channel can be moved any
time. Right mouse click removes control points. The transfer function can be
reset to identity using the “Reset” button. The “Save table” button saves a
created transfer function in color table (.ct) format. The “Load table” button
opens a file dialog to load previously saved transfer functions (color tables).
Blending On the bottom of the transfer function editor, a spinbox allows to
choose a blend mode. Default is “normal” blending, adjusting transparency to
alpha value.
Alternative blend modes are MIP (maximum intensity projection) and X-ray. The
first one maps the pixel with the largest gray value (the brightest pixel) to the
destination pixel, the latter accumulates the values of successive pixels,
generating an X-ray like visualization. Both methods create very bright, highly
oversaturated images. The saturation can be scaled by reducing the “Blend
scale” with the slider on the bottom of the transfer function editor.
Isosurface editor The isosurface editor allows to define a threshold value to
set pixels to totally transparent (A=0). You can choose if the values below,
above, or exactlythe same as the threshold should be set to transparent by
clicking on one of the three radio buttons below the function widget.
This is useful if the object in your image has a fuzzy environment which should
not be displayed. It is also possible to set these black environment pixels to
transparent by using the transfer function editor and modifying the alpha
channel. But this is the faster method.
Clipping editor The clipping editor enables a clipping plane cutting the
image and providing a view of the interior of your image, see Figure 50(a). With
the three scrollwheels it is possible to rotate the plane around all three axes. If
MAVI user manual
102
Volume Rendering View
the “View aligned clipping plane” button is checked, the clipping plane always
faces towards the view direction.
To change the clipping direction, you can either
(1)
(2)
rotate the object using the
button in the toolbar (Figure 50(c)) or
deactivate the “View aligned clipping plane” button and choose the
plane’s direction using the X, Y, Z axis wheels.
Note that activating the “View aligned clipping plane” button again changes
back, the chosen plane direction is reset.
To change the clipping plane’s position w. r. t. the image, move the cutting
plane through the image with the position slider at the bottom (Figure 50(d)
and (e)).
Note that this position slider can be used to generate frames for a movie
showing the successive cutting of the structure, too.
Light editor Enable lighting to emphasize the 3D effect. Color nuances can
be enhanced by modification of the light color. Showing the light position can
help fine-tuning. The threshold controls the degree to which the light is
absorbed by the structure within the image.
View parameter editor Determine the position of the object to be rendered
setting the “Rotate Local” and “Rotate Global” parameters, see Figure 51(a).
Useful to recover exactly the same view angle when renderings of two different
images shall be comparable.
The “Render Window”parameters control the size of the 2D image holding the
rendering result if it is saved, see Figure 51(b).
Preferences editor You can customize the rendering process and
enable/disable debugging information in the preferences editor:
Draw bounding box: enables/disables the display of the yellow bounding box
and the green direction indicator
Draw scale: enables/disables the display of the along the axes of the bounding
box
Draw red-green: enables a red-green pseudo 3d rendering mode which must
be viewed with red-green goggles. Note that the rendering performance will
be lower in this mode, because the image has to be drawn twice in one
render pass.
MAVI user manual
103
Volume Rendering View Selection
Draw proxy geometry: disables texture application and draws the underlying
geometry used by the render algorithm
Enable progressive refinement: temporarily shrinks the image when
rotating
Set Background Color: opens a color editor window to choose a new
background color
Set Boundingbox Color: opens a color editor window to choose a new color
for the bounding box
Set Label Color: opens a color editor window to choose a new color for the
axes labels 0, x, y, z, and the scale limits
At the bottom of the editor widget, there is another tab, which shows a list of
all the OpenGL extensions your graphic card supports. This is useful for
debugging purposes but there is nothing to adjust here. The best rendering
procedure for your graphic card is chosen automatically.
8.6.4
Volume Rendering View - Selection
This function offers a choice of render engines to be employed in the rendering
process. Before opening the Volume Rendering View, a dialog showing the
render engines available for your graphics card is opened.
The 3D_T_CG render engine only uses memory of the same size as the image
data. It is the best choice if light effects are nor required.
For GRAY8 and label images, Pre-Classification or Post-Classification can be
chosen. Pre-Classification means that the RGBA value of a pixel in the
visualization is obtained by first applying the transfer function, which assigns
RGBA color values to the image pixels, and then interpolating the resulting
RGBA values of the surrounding image pixels. With Post-Classification, the gray
value of the visualization pixel is interpolated from the gray values of the
surrounding image pixels and then the transfer function is applied to obtain the
corresponding RGBA value.
Usually, post-classification creates more realistic, smooth renderings. However,
for label images, post-classification causes artifacts along the borders of the
labeled objects due to the intermediate colors obtained by interpolation, see
Figure 52. Thus pre-classification should be chosen in cases where crisp jumps
between colors are desired.
MAVI user manual
104
Volume Rendering View Selection
(a) Start
(b) Change background color, adapt transfer function
(c) Change blend scale
(d) Add light
(e) Fine tune light
Figure 48: Step-by-step volume rendering.
MAVI user manual
105
Volume Rendering View Selection
(a) 0
(b) 35
(c) 58
(d) 135
Figure 49: Example effect of isosurface threshold. Pixels with gray value below
the threshold are completely transparent.
MAVI user manual
106
Volume Rendering View Selection
(a) Enable clipping
(b) View aligned clipping
(c) Rotate sample
(d) Rotate sample
(e) Move plane
(f) Move plane
Figure 50: Volume Rendering View: enabling and positioning of clipping
plane.
MAVI user manual
107
Volume Rendering View Selection
(a) Rotate parameters
(b) Render window
Figure 51: Volume Rendering View view parameter settings.
MAVI user manual
108
Volume Rendering View Selection
(a) Pre-classification
(b) Post-classification
Figure 52: Volume Rendering View, effect of choosing pre- or postclassification when rendering a label image. Post-classification generates unwanted grayish color artifacts at the surfaces of the spheres.
MAVI user manual
109
Help
8.7
Help
8.7.1
System Info
Pops up a window with essential information on
• operating system (OS)
• processor(s) (CPU)
• memory (RAM)
• graphics card (GPU).
Please provide this information e. g. via a screen shot when asking for e-mail
support at [email protected]
8.7.2
Help Assistant
The Help Assistant provides the complete MAVI user documentation handbook
(this document) online. Clicking the respective button opens a separate window
enabling navigation via content tree and index as well as a key word search.
8.7.3
About
Pops up a window with essential information on the version of MAVI you’re
running. Please provide the full information in the “Version Info” tab e. g. via a
screen shot when asking for e-mail support at [email protected]
MAVI user manual
110
Frequently Asked Questions
9
Frequently Asked Questions
Where do I report problems with MAVI? Please report all bugs, problems,
and feature wishes to [email protected] Please always attach the
Version Information from the plugin-tab in the Help:About dialog, and the
logfile mavi.log in \$HOME/.ITWM/ (linux) or ApplicationData\ITWM (windows).
The slice view is black, but there should be objects/structure. If your
image is a labeled one, obtained by labeling or the watershed transformation,
try setting the color filter manually. If your image is of type GRAY16, try using
the 12 Bit Gray Scale view filter. More general, Manipulation:Equalize the
gray value histogram.
I tested the FieldFeatures/ObjectFeatures using a cube parallel to the
coordinate directions. Why is the surface area measured by MAVI too
small? This is due to MAVI minimizing the mean error in general position.
See the Remark at the end of Section 8.5.8 for details.
I measured both the FieldFeatures and the ObjectFeatures for my MONO
image. Why do the results for the surface area and the integral of mean
curvature differ? This is due to the different treatment of edge effects. See
the Remark at the end of Section 8.5.8 for details.
When I’m connected to a machine running SUSE Linux Enterprise 11
using ssh, MAVI crashes with a segmentation fault. This is an OpenGL
library/driver problem on machines running SUSE Linux Enterprise 11 with
servicepack 0 and servicepack 1. It is fixed with servicepack 2, so please update
the remote machine to servicepack 2. An easy workaround if you can not
update: set LANG to en_ENutf8.
MAVI user manual
111
Optional modules
10
Optional modules
The following modules are not part of MAVI under the standard full license.
They can be added on user’s demand. In that case, they are integrated
completely within MAVI’s framework.
10.1
Particle Features
This module adds analysis options for image objects exceeding the range of
functionality of the Object Features. As the latter, the Particle Features work
on label images as generated e. g. by Labeling, Particle Separation or Cell
Reconstruction. The following characteristics are provided by the Particle
Features:
• length, width, and thickness derived from the sorted edge lengths of the
minimal volume bounding box (cuboid)
• maximal diameter
• maximal local thickness
• elongation and elongation index.
The results can be exported as CSV.
Combined with the Object Features, these features enable easy classification
of particles into the classes grain, chip, and fiber. See [23] for details.
10.2
Point Field Statistics
On the basis of the Object Features, this module offers additionally functions
for analyzing the spatial arrangement of the found objects. To this end
summary statistics for random point fields from spatial statistics are computed
for the object centers. Suppose the object center’s to form a realization of the
random point field Φ with distribution P and expected point density λ.
The statistics provided in the module are
• Ripley’s K-function K(r) (mean number of r-close pairs of points)
X
1
K(r) =
Expectation
Φ ∩ Ball(x, r) \ {x}
λVolume(B)
x∈Φ∩B
• L-function L(r) (normalized version of K)
s
L(r) =
3
K(r)
Volume(Ball(o, r))
MAVI user manual
112
Point Field Statistics
• empty space function, F (r) (distribution function of the distance of an
arbitrary point to the next object center)
F (r) = P (Φ ∩ Ball(o, r) \ {0} =
6 ∅)
• nearest neighbor distance distribution function , G-function G(r)
(distribution function of the distance of an object center to the next object
center)
G(r) = P (Φ ∩ Ball(o, r) \ {0} =
6 ∅|0 ∈ Φ)
• J-function J(r) = (1 − G(r))/(1 − F (r))
• paircorrelation function, g(r) (roughly the probability density of K)
The results can be exported as CSV.
The points of a Poisson point field are independently identically uniform
distributed without interaction among each other. Their number is Poisson
distributed. The Poisson point field often serves to compare observed point
patterns to it. If there is interaction between the points, one usually describes it
as clustering (in case of attraction) or regularity (in case of repulsion).
For 2D Poisson point fields it holds that
K(r)
L(r)
F (r)
G(r)
g(r)
=
=
=
=
=
πr2
r
1 − exp(−λπr2 )
F (r) and thus J(r) = 1
1
For 3D Poisson point fields it holds that
4 3
πr
3
L(r) = r
K(r) =
4
F (r) = 1 − exp(− λπr3 )
3
G(r) = F (r) and thus J(r) = 1
g(r) = 1
In a Poisson point field, the points are independent, thus counting neighbors of
a point of the field is just the same as counting neighbors of an arbitrary point.
Hence G = F and J = 1.
In a clustered point field, a point of the field has more close neighbors than an
arbitrary one. Thus G > F and J < 1.
In a regular point field, a point of the field has less close neighbors than an
arbitrary one. Thus G < F and J > 1.
MAVI user manual
113
Point Field Statistics
Note however, that a point field can be both clustered and regular. (Think of
clusters centered at the vertices of a grid or small, grid-like structures forming
larger clusters.) Therefore, the range or distance r has to be considered, too.
Table 20 below summarizes the rules of thumb:
Point field type
Poisson
Regular
Clustered
F, G
F =G
F >G
F <G
J
J =1
J >1
J <1
g
g=1
g<1
g>1
Table 20: Point field statistics indicating clustering, regularity, or independence.
Figure 53 shows realizations Poisson point fields and typical examples of a
clustered and a regular point field. Figure 54 provides the summary statistics for
these realizations.
(a) Poisson, 2D
(b) Matérn Cluster, 2D
(c) SSI, 2D
(d) Poisson, 3D
(e) Matérn Cluster, 3D
(f) SSI, 3D
Figure 53: Realizations of random point fields in 2D and 3D. Matérn Cluster is
clustered, SSI is regular in the sense that the points observe a minimal distance.
See [4] for more mathematical background.
MAVI user manual
114
Mesh Export
(a) K-functions 2D
(b) K-functions 3D
(c) F-functions 2D
(d) F-functions 3D
(e) G-functions 2D
(f) G-functions 3D
(g) J-functions 2D
(h) J-functions 3D
(i) g-functions 2D
(j) g-functions 3D
Figure 54: Summary statistics for the point field realizations shown in Figure 53.
10.3
Mesh Export
For segmented MONO images, this module offers surface meshing, mesh
simplification, and export.
Meshing generates a representation of the image data by a triangulation of the
surface of the foreground. The original mesh is based on the adjacency system
and subdivides each surface pixel into several triangles.
This initial mesh has to be simplified. Mesh Export unites nearly co-planar
triangles to a user defined proportion of the initial triangle number while
preserving the topology of the foreground.
Finally, the mesh is exported as STL file (binary, can be compressed as GZ).
MAVI user manual
115
Mesh Vis
10.4
Mesh Vis
This function adds interactive visualization to the functionality of Mesh Export.
It can only be used together with Mesh Export and offers:
• Import a surface mesh in STL format (binary or ASCII, can be compressed as
GZ)
• Visualize as surface mesh, edge mesh, or point cloud
• Rotate, Zoom, Move
• Rotate automatically, adjustable tracking shot
• Simplify the mesh in individual steps, to a user defined number of triangles,
or to a user defined degree of co-planarity
• Smooth the mesh
• Export the current view (screenshot)
• Export automatic rotation or tracking shot as key frames
10.5
Gray Value Mapping
In computed tomography images high gray values correspond to a high
(physical) density of the material. The Gray Value Mapping helps to check the
homogeneity of a material: density variations like inclusions or condensations
can be visualized.
The plugin offers a representation with iso lines, where the iso levels correspond
to the deviation from the mean density. It is possible to specify the size of the
filter mask and a mask image (to exclude background or certain inclusions). The
view and the palette can be saved.
The filter mask should be chosen slightly bigger than the diameter of negligible
structures in the image, such as small inclusions, that you do not want to
visualize.
MAVI user manual
116
Gray Value Mapping
Figure 55: Gray Value Mapping window.
MAVI user manual
117
REFERENCES
References
[1] J. Canny. A computational approach to edge detection. IEEE Trans. Pattern
Analysis and Machine Intelligence, 6(PAMI-8):679–698, 1986.
[2] M. Frigo and S. G. Johnson. FFTW 2.1.3 (the fastest Fourier transform of
the West). http://www.fftw.org/, 1998.
[3] M. Frigo and S. G. Johnson. FFTW: An adaptive software architecture for
the fft. In Proceedings of the International Conference on Acoustics,
Speech, and Signal Processing, volume 3, pages 1381–1384, May 1998.
[4] J. Illian, A. Penttinen, H. Stoyan, and D. Stoyan. Statistical Analysis and
Modelling of Spatial Point Patterns. Statistics in Practice. Wiley, Chichester,
2008.
[5] K. Koch, J. Ohser, and K. Schladitz. Spectral theory for random closed sets
and estimating the covariance via frequency space. Adv. Appl. Prob.,
35:603–613, 2003.
[6] C. Lang, J. Ohser, and R. Hilfer. On the analysis of spatial binary images.
Journal of Microscopy, 203:303–313, 2001.
[7] C. R. Maurer and V. Raghavan. A linear time algorithm for computing
exact Euclidean distance transforms of binary images in arbitrary
dimensions. IEEE Trans. Pattern Analysis and Machine Intelligence,
25(2):265–270, 2003.
[8] W. Niblack. An introduction to digital image processing. Prentice-Hall,
1986.
[9] M. Nöthe, K. Pischang, P. Ponižil, B. Kieback, and J. Ohser. Study of
particle rearrangements during sintering processes by microfocus
computer tomography (µct). In Proc. World Congress PM2004 Powder
Metallurgy, Vienna, October 17th-21st 2004.
[10] J. Ohser, W. Nagel, and K. Schladitz. Miles formulae for Boolean models
observed on lattices. Image Anal. Stereol., 28(2):77–92, 2009.
[11] J. Ohser, C. Redenbach, and K. Schladitz. Mesh free estimation of the
structure model index. Image analysis and stereology, 28(3):179–186,
2009.
[12] J. Ohser and K. Schladitz. 3d Images of Materials Structures – Processing
and Analysis. Wiley VCH, Weinheim, 2009.
[13] S. Osher and L. Rudin. Feature-oriented image enhancement using shock
filters. SIAM J. Numer. Anal., 7:919–940, 1990.
[14] A. Rack, L. Helfen, T. Baumbach, S. Kirste, J. Banhart, K. Schladitz, and
J. Ohser. Analysis of spatial cross-correlations in multi-constituent volume
data. Journal of Microscopy, 232(2):282–292, 2008.
[15] C. Redenbach, A. Rack, K. Schladitz, O. Wirjadi, and M. Godehardt.
Beyond imaging: on the quantitative analysis of tomographic volume data.
Int J Mater Res, 2:217–227, 2012.
[16] J. Sauvola and M. Pietikäinen. Adaptive document image binarization.
MAVI user manual
118
REFERENCES
Pattern Recognition, 33:225–236, 2000.
[17] K. Schladitz, C. Redenbach, T. Sych, and M. Godehardt. Model based
estimation of geometric characteristics of open foams. Methodology and
Computing in Applied Probability, 2010. Accepted.
[18] K. Schladitz, C. Redenbach, T. Sych, and M. Godehardt. Model based
estimation of geometric characteristics of open foams. Methodology and
Computing in Applied Probability, pages 1–22, 2011.
[19] P. Soille. Morphological Image Analysis. Springer, Berlin, Heidelberg, New
York, 1999.
[20] D. Stoyan, W. S. Kendall, and J. Mecke. Stochastic Geometry and Its
Applications. Wiley, Chichester, 2nd edition, 1995.
[21] O. D. Trier and A. K. Jain. Goal-directed evaluation of binarization
methods. IEEE Transactions on Pattern Analysis and Machine Intelligence,
17(12):1191–1201, 1995.
[22] Y. F. Tsao and K. S. Fu. A parallel thinning algorithm for 3-d pictures.
CVGIP, 17:315–331, 1981.
[23] I. Vecchio, K. Schladitz, G. Godehardt, and M. Heneka. 3d geometric
characterization of particles applied to technical cleanliness. Image
Analysis and Stereology, 31(3):163–174, 2012.
[24] L. Vincent and P. Soille. Watersheds in digital spaces: An efficient
algorithm based on immersion simulation. IEEE Transactions on Pattern
Analysis and Machine Intelligence, 13(6):583–598, 1991.
[25] P. Zamperoni. Methoden der digitalen Bildsignalverarbeitung. Vieweg,
Braunschweig, 1991.
MAVI user manual
119
Index
(Sub-)Field Fiber Directions, 89
About, 110
Absolute Difference, 43
Adaptive H-Extrema, 66
Add, 43
Add Value, 42
adjacency system, 7
Algebraic Closure, 60
Algebraic Opening, 59
And, 43
anisotropy, 79
degree of, 91
Approx. Ball, 56
area
density, 81
of an object, 86
of foreground, 81
of projections, 80
total, 81
Area Fraction Profile, 83
Autocovariance, 93
(Rotation Mean), 94
AutoCrop, 36
avs data import, 28
Bartlett Spectrum, 94
(Rotation Mean), 94
From Autocovariance, 94
Binarization, 44
binarization, 44
Binary Operations, 43
Black Top Hat, 59
boundary length, 81
density, 81
of an object, 86
Cast, 37
Cell Reconstruction, 54
by complex morphology, 54
by preflooded watershed, 54
by smoothing, 54
center
of an object, 85, 86
close, 33
all documents, 33
document, 33
view, 33
Closure, 59
closure
algebraic, 60
color table, 101
Complement, 39
Complex Conjugate, 40
connected components, 47
connectivity, 7
discrete, 7
Convert, 37
Convolution, 93
coordinate system, 7
Covariance, 93
(Rotation Mean), 93
Crop, 34
crop
automatically, 36
CSV file, 31
Cut Hills, 72
cylindrical sample, 41
demo version, 2
diameter
of an object, 85, 86
of equal area circle, 86
dicom data import, 28
Dilation, 58
directions
discrete, 8
120
INDEX
discrete normal, 9
Discrete Distance Transformation, 63
distance transformation
L1 , 63
L∞ , 63
Chamfer, 63
discrete, 63
Euclidean, 62, 63
Manhattan, 63
maximum, 63
w.r.t. adjacency system, 63
Divide by Value, 42
double thresholding, 45
dynamic, 65
dynamic range, 37
edge
correction, 87
effect, 87
edge treatment
embedded, 10
no, 10
periodic, 10
reflective, 10
reflective double edge, 10
edit header, 33
EMB, 10
embedded, 10
empty space function, 112
Equalize, 38
Erosion, 57
Euclidean Distance Transformation, 62
Euclidean Distance Transformation
(Voronoi algorithm), 63
Euler number, 80, 81
density, 80, 81
of an object, 85, 86
export, 31
analysis data, 31
CSV files, 31
image data, 31
Extended H-Maxima, 66
Extended H-Minima, 66
Extract by label, 36
Extract Hills, 72
Extract Holes, 72
extract object, 36
extrema
regional, 65
Fast Fourier Transformation, 71
inverse, 71
fiber
orientation, 89
tutorial, 22
reinforced composites, 89
Field Features, 79
precision of, 82
file
iass, 27
iass.gz, 27
STL, 113
Fill Holes, 72
Filter
Alpha-Trimmed Mean, 57
Binomial, 57
Gauss, 57
High Pass, 57
Laplace, 57
Mean, 57
Median, 57
Weighted Mean, 57
filter, 57
edge detection, 57
morphological mean, 61
morphological shock, 61
nonlinear, 57
smoothing, 57
view, 100
foam
cell reconstruction, 54
open cell
mean characteristics, 88
Frequently Asked Questions, 111
G-function, 112
Gaussian curvature, 86
geodesic
reconstruction, 68, 69
GeoDict volume data import, 31
Geometric Tortuosity, 89
global thresholding, 44
MAVI user manual
121
INDEX
Granulometry, 84
granulometry
spherical, 84
gray value
histogram, 78
rotation mean, 78
type
change, 37, 38
types, 7
Gray Value Mapping, 114
Grayvalue Profile, 83
H-Concave, 66
H-Convex, 66
H-Extrema, 65
adaptive, 66
H-Maxima, 65
H-Minima, 65
Help Assistant, 110
hills
cut, 72
extract, 72
histogram, 78
history, 6
holes
extract, 72
fill, 72
homogeneity, 82, 83, 114
Hysteresis thresholding, 45
iass file, 27
image, 6
attributes, 6
view, 98
COMPLEX, 7
creator, 6
description, 6
GRAY16, 7
GRAY32, 7
GRAY8, 7
header, 6
edit, 33
set entries, 33
MONO, 7
size
change, 34, 36, 37
decrease, 34, 37
increase, 36
type, 7
change, 37, 38
Image Attributes View, 98
Image Statistics, 78
import
3D electron density files, 31
avs data, 28
dicom data, 28
Fraunhofer rek files, 31
GeoDict volume data, 31
image raw data, 28
image stacks, 30
mrc, 31
integral of curvature
of an object, 86
integral of mean curvature, 80
density, 80
fractal dimension, 80
of an object, 85
integral of total curvature, 80, 81
interface, 2
intrinsic volumes, 79
Inverse Fast Fourier Transformation, 71
invert, 39
J-function, 112
K-function, 112
L-function, 112
label image, 47
extract object from, 36
prepare for visualization, 39
remove objects from, 74
Labeling, 47
labeling
by reconstruction, 47
lattice
anisotropic, 6
isotropic, 6
lattice spacing, 6
lattice spacings
correct, 33
length
of an object, 112
MAVI user manual
122
INDEX
lengths of projections, 80
local
pore size, 84
thickness, 84
thresholding, 47
local thresholding, 46
Magnitude
of Fast Fourier transformation, 71
main menu
Analysis, 78
Spectral Methods, 93
File, 27
Export, 31
Import, 28
Help, 110
Manipulation, 34
Binary Operations, 43
Segmentation, 44
Cell Reconstruction, 54
Particle Separation, 55
Transformation, 56
Distance Transformation, 62
Filter, 56
Geodesic Transformation, 65
Morphology, 56
Spectral Transformation, 71
View, 95
main window, 2
Map Subfield Feature To Image, 74
Map Subfield Fiber Directions To
Image, 75
Mask Image, 40
Mask with Cylinder, 41
material density, 114
maxima
regional, 65, 72
Maxima Imposition, 67
Maximum, 43
mean chord length
of an object, 86
mean chord lengths, 80, 81
medial axis, 72
Mesh
Export, 113
Vis, 113
mesh
simplification, 113
visualization, 113
meshing, 113
MIL, 80, 81
Miles-Lantuejoul, 87
correction, 87
weight, 87
minima
regional, 65, 72
Minima Imposition, 66
minimal volume bounding box, 112
Minimum, 43
Minkowski functionals, 79
morphological
black top hat, 59
closure, 59
dilation, 58
erosion, 57
opening, 58
white top hat, 59
Morphological Gradient, 61
Morphological Mean, 61
morphology
geodesic, 65
Multiply, 43
Multiply by Value, 42
nearest neighbor distance distribution
function, 112
neighborhood, 7
discrete, 7
graph, 7
NET, 10
Niblack Segmentation, 46
no edge treatment, 10
object
bounding box, 85
center, 85
diameter, 85
diameter of equal volume ball, 86
labeling, 47
Object Features, 85
Object Filter, 74
offset
MAVI user manual
123
INDEX
of an object, 85, 86
of image, 6
Open Foam Features, 88
Open Image, 27
Opening, 58
opening
algebraic, 59
curve, 84
Or, 43
orientation
distribution of fibers, 89
tensor, 90
orientation analysis
tutorial, 22
orientation tensor
tutorial, 22
Otsu’s threshold, 45
Pad, 36
paircorrelation function, 112
particle
separation
tutorial, 24
Particle Features, 112
Particle Separation, 55
by complex morphology, 55
by preflooded watershed, 55
by smoothing, 55
pattern sectrum, 84
PER, 10
periodic, 10
Phase
of Fast Fourier transformation, 71
pixel, 6
gray value, 7
pixel spacing, 6
pixel spacings
correct, 33
Point Field Statistics, 112
pore size
local, 84
post-classification, 104
Power Spectrum, 71
ppi value, 89
pre-classification, 104
preferred directions, 79
Prepare label image for visualization,
39
quermass integrals, 79
quit, 33
raw data import, 28
RDE, 10
Reconstruction by Dilation, 69
Reconstruction by Erosion, 68
REFL, 10
reflective, 10
reflective double edge, 10
Regional Extrema, 65
render engine, 104
Rotation Mean, 78
rotation mean
of autocovariance, 94
of Bartlett spectrum, 94
of covariance, 93
Sauvola Segmentation, 47
save
slice, 27
Save Image, 27
Self Dual Reconstruction, 69
separation
of particles
tutorial, 24
shading correction, 22
shape factor
isoperimetric, 87
Shrink, 37
size
of an object, 85, 86
Skeleton, 72
Analyzer, 73
skeleton
bisector function, 73
classify pixels, 73
medial axis, 72
Slice View, 95
SMI, 80
spacing, 6
spacings
correct, 33
spectral transformations, 71
MAVI user manual
124
INDEX
spectrum
Bartlett, 94
pattern, 84
power, 71
sphericity, 87
Spread, 37
stack import, 30
status
of an object, 85
STL, 113
structure model index, 80
structuring element, 56
Approx. Ball, 56
subfield, 89
Subfield Features, 82
Subtract, 43
Subtract Value, 42
support, 111
surface
area, 80
density, 80
fractal dimension, 80
surface area
of an object, 85
of convex hull of an object, 85
System Info, 110
thickness
local, 84
of an object, 112
thinning, 72
topology preserving, 72
thresholding, 44
double, 45
global, 44
hysteresis, 45
local, 46, 47
Niblack, 46
Sauvola, 47
Toggle Mapping Erosion/Dilation, 61
Toggle Mapping Opening/Closure, 62
topological core, 73
tortuosity
geometric, 89
total curvature density, 80, 81
transfer function, 100
triangulation, 113
tutorial
adaptive h extrema, 20
cell reconstruction, 20
closed foam, 20
fiber orientation, 22
fiber reinforced composite, 22
mean cell size, 16
open foam, 16
orientation tensor, 22
particle separation, 24
porosity, 16
two-point correlation function, 93
Ultimate Dilated Set, 68
Ultimate Eroded Set, 67
Unary Operations, 42
view
filter, 100
image attributes, 98
slice, 95
active window, 95
volume rendering, 98
blending, 102
clipping, 102
isosurface, 102
light, 103
transfer function, 100
volume
density, 80
fraction, 80
of an object, 85
of convex hull of an object, 85
of foreground, 80
total, 80
volume data import, 28, 30
Volume Rendering View, 98
Selection, 104
voxel, 6
Watershed Transformation, 47
Preflooded, 48
White Top Hat, 59
width
of an object, 112
workflow
MAVI user manual
125
INDEX
first, 13
typical, 13
MAVI user manual
126
Was this manual useful for you? yes no
Thank you for your participation!

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

Download PDF

advertisement