Introduction to Quantitative

Image Analysis
Karl-Heinz Wolf, Joost van Meel
A course for MSc-students Geoscience & Engineering
Learn how to recognize and measure rock textures and
specific texture features.
Learn how to deal with pitfalls in image manipulation.
Basics for 2D 3D, i.e. slices to CT-scan volumes.
WHY?
Content Day 1
Sources of images
Digitizing images
Analysis of an image
Binary image
Processing of images
*Point Processing
*Neighborhood
Processing
More terminology
Content Day 2
Special Operations
*Logical Operators
*Skeleton
*Pruning
*Skiz
Content day 3: Excercises
Sources of Images
Image of a photocamera / a (digital) videocamera
Microscopy Images (camera on microscope)
Electron Microscopy Images, CT-scans
Images (frames) from a video movie
Electronic information from sensors, arranged in a
2D matrix. These sensors may record electronic or
magnetic signals
Sour c es & Ex ampl e of SEM
Cor e sampl e
Mi c r ophot o
Thi n sec t i on
2-D //-ni c ol s
Thi n sec t i on
2-D X-ni c ol s
Sour c es & Ex ampl e of Thi n Sec t i ons
Sour c es & Ex ampl e of CT i mages of por es
& gr ai ns
Bent hei m sst c or e
X-Y CT sc an and
X-Z r ec onst r uc t i on
x
y
z
CT-sc an r ec onst r uc t i on of gr ai ns
Por e body r ec onst r uc t i on
Sour c es & Ex ampl e of CT i mages of Fr ac s
Or i ent at i on
Open f r ac t ur es
Cement ed
f r ac t ur es
Repeating Image Pattern
Feret :
(breadth,
length)
Roundnes,
aspect ratio
direction of an
object
This course is about obtaining quantitative data
from digitized images
1 2 3
Counting Objects
Intergrowths
Digitizing an Image-1
If an image is digitized, it becomes subdivided Picture
Elements (pixels)
Each pixel is valued in ones and zeros
The brightness of a pixel is subdivided in 8 ones and zeros;
8 bits = 1 byte
This results in 256 grey values, varying from
00000000 (black) = 0 to 11111111 (white) = 255
With a green, blue and red component, represented by three
256 grey values, the colour spectrum is represented with
256
3
= 18609625 colours.
Digitizing an Image -2
A digitized image is a 2D matrix of pixels. In this matrix,
for every pixel a code is present which indicates the
greyvalue of the pixel
Image with 2 magnifications. In the
last one, one may see the pixels.
Analysis of an Image
The properties of an image may be recorded on two levels :
Grey / Colour Level
Measurements are conducted directly on the image, for
instance to determine the brightness distribution in an
image
Binary Level
Here a process is used that is termed detection, or
thresholding. Pixels matching a certain criteriumare
recorded in a specific part of the memory of the computer :
a binary image
Binary image -1
Using detection, or thresholding it is investigated if a pixel
does or does not match a certain criterion (for instance a
specific grey value). Does the pixel match the criterion, then
the pixel is set on (1). If not, the pixel is not set (0).
The result is called binary, because the pixels in the resulting
image (present in another binary image) are either on (value
1) or off (value 0).
Therefore a binary image has NO grey levels !
Binary image -2
Detection
(criterion)
:
white
Binary Image
Processing of Images -1
For the processing of images there are generally
two different ways :
Point processing
Neighbourhood Processing
Processing of Images -2
Point Processing
In this type of processing, the value of a pixel is
changed without taking into account the value of
neighboring pixels.
Examples of this are:
- operations where brightness or contrast are
changed, or,
- an operation called detection (thresholding).
Point Processing-1
+10
+10
Brightness
Brightness
0
0
255
0
10
255
10
10
255
10
20
255
Note: 00000000 (black) = 0 to 11111111 (white) = 255
Grey Image
Grey Image
Point Processing-2
x10
x10
0
0
255
0
0
255
0
10
255
0
100
255
Contrast
Contrast
Grey Image
Grey Image
Point Processing-3
Threshold
> 25
0
50
255
Detection or
Thresholding
0
1
Grey Image Binary Image
Neighbourhood Processing-1
This can be subdivided into :
Mathematical Morphology
Examples : Erosion, Dilation, Sharpen, Smoothing,
Edge Detection
Convolution
Examples : Gauss Filters, Sobel Filters
Neighbourhood Processing-2
Structuring Element :
also called the operator. It is a square matrix of pixels
moving over the object.
The Structuring Element defines the behaviour of erosion
and dilation
To erode an object, every pixel must be compared with its
neighbours.
Neighborhood Processing-3
In a binary image every pixel can be turned on (1) or off
(0). The structuring element moves over the object. The
value of the central pixel p becomes dependent on the
neighboring pixels
If a pixel in a binary image coincides with the boundary of
the object, at least one of the neighboring pixels will have the
value zero (0). In the erosion process, the central pixel is set to
zero (0) too.
p
Structuring element
Neighbourhood Processing - 4
Erosion :
this proces erodes a layer of pixels around an object. Only
those pixels are removed, which have a boundary with the
background.
It leads to a different result if one takes into account only
the sides of the pixels (4-connected erosion) or sides and
corners (8-connected erosion).
After erosion of a pixel, the original image is taken as source
for the next erosion step (non- iterative process).
Neighbourhood Processing-
Erosion - animation
P
An object and
the structuring element
Neighbourhood Processing-
Erosion
Red = unset, Green = set
Note: The pixel matching the central
pixel will be marked, and later be unset
Erosion : If at least one pixel in the
structuring element coincides with a
pixel which is unset, the pixel matching
the central one will be unset too. If the
central pixel does not coincide with a
pixel of the object, nothing happens
P
Neighbourhood Processing-
Erosion
P
Neighbourhood Processing-
Erosion
Pixels marked will be unset,
when all pixels of object
have been investigated
P
Neighbourhood Processing-
Erosion
P
Neighbourhood Processing-
Erosion
P
Neighbourhood Processing-
Erosion
P
Neighbourhood Processing-
Erosion
P
Neighbourhood Processing-
Erosion
P
Neighbourhood Processing-
Erosion
P
Neighbourhood Processing-
Erosion
P
And so on......
Neighbourhood Processing-
Erosion
Neighborhood Processing-
Erosion
After a while .....
P
Neighbourhood Processing-
Erosion
After another while .....
Orange pixels will be eroded in
8-connection erosion, but not in
4-connection erosion
Neighbourhood Processing-
Erosion
Final state of
4-connected erosion
Marked pixels will be deleted
Neighbourhood Processing-
Erosion
Final state of
8-connected erosion
Marked pixels will be deleted
This final state is also used in
Edge Detection (contouring)
Neighbourhood Processing - 5
Dilation : this is exactly the opposite of erosion: a layer of
pixels is added to the object.
N.B. : In general dilation after erosion will ONLY
give the original object when it is a circle!!
Neighbourhood Processing - 6
Open: an erosion followed by a dilation
Close: a dilation followed by a erosion
Neighbourhood Processing - 7
In a grey level image the central pixel p will be set to the
minimum value of the neighbouring pixels.
Edge detection in binary image : if one of the neighbouring
pixels has the value zero, the central pixel coincides with a
pixel on the boundary of the object. If the central pixels
matching this criterion are recorded, this results in another
binary image containing only the edge (contour) of the
object.
Erosion Dilation
Exercise
8-connected processes
1. Erosion of the objects
2. Dilation of the remaining
shapes
More Terminology
More Terminology-1
Feature Count Point (FCP): the coordinates of a
specific pixel, recorded by the program as being
the location of the object. In the software used in
the course this is the lowest rightmost pixel of the
object. Other software packages may use another
pixel for this .
Image Frame: the frame indicating which part of
the image should be processed (red frame)
Measure Frame: the frame indicating which part of
the image should be measured (blue frame).
N. B. : Objects with their FCP outside the measure
frame are ignored in measurement !!!!
More Terminology-2
Feret : measure of length in a certain direction
the length of the object is the longest feret,
de breadth the shortest feret
Feret : think of Calliper
embayments are not taken
into account
Breadth
Length
and
orientation
More Terminology-3
Convex perimeter: The length of the
polygon fitting to the object
Think of a rubber band around the object.
Convex area: the area of the mentioned
polygon.
Area: area of the object
Orientation: orientation of the longest feret
Aspect ratio: ratio of shortest and longest feret
More Terminology-4
Perimeter: real perimeter, or the total length of the
boundary of the detected object
Roundness: a shape factor defined as
064 . 1 * * * 4
2
area
perimeter
roundness

More Terminology-5
In the roundness equation, what is the factor 1.064 for?
The perimeter is nothing more but the sum of all such pixels.
This is different from the true length of a curve.
The factor 1.064 is an empirical correction factor for this effect.
Program: Leica QWIN
Image files:
Group (G: ) : .AES0101
File names probably without extension !
DAY 2
More Operations!
Logical Operators-1
Because binary images consist of pixels with value 1 or 0, it is
possible to use Boolean logical rules for adding and
subtracting the images .
The logical condition NOT must be interpreted as
INVERTED : every pixel which is set becomes unset, and
vice versa
Logical Operators-2
Skeleton
Skeletonizing is a special kind of erosion, creating a
conditional thinning of the features in an image.
Exhaustive Skeletonizing reduces features to a single pixel
width, thus thinning the feature to its central framework
Pruning-1
Very often skeltonized features contain many branches or
dendrites, or they may be caused by the digital nature of the
pixel grid.
To reduce these effects, the skeleton can be pruned by
iteratively removing the end points.
Pruning is a conditional erosion, in which only those pixels are
removed, that have only one side connected to their neighbour.
Pruning can also be exhaustive, in which no branches are left,
only loops
Pruning-2
Image
Skeleton
Pruned skeleton
(non-exhaustive)
Skiz-1
Skiz: Abbreviation for Skeleton Influence Zone
It is a combination of operations :
1. inversion of the image
2. exhaustive skeleton of the inverted image
3. exhaustive pruning of the skeleton
Resulting from Skiz are boundary lines between features,
which separate the influence zone of pixels of a feature
Skiz-2
A group of features and the
boundary lines formed by
the Skiz operation
The dark spots are the
original features, the lines
represent the boundaries of
equal distance.
Assignments