Download - DIP-5 en ImageAnalysis Part2-Very Nice
![Page 2: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/2.jpg)
Département GE - TI - Olivier Bernard 2
Summary
I. IntroductionII. Digital image fundamentalsIII. Discrete 2D processingIV. Image improvement
V. Image analysis (6 houres)
![Page 3: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/3.jpg)
Département GE - TI - Olivier Bernard 3
Image analysis
Gonzales : Digital Image Processing, Prentice Hall (3th Ed.)
� Books
![Page 4: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/4.jpg)
Département GE - TI - Olivier Bernard 4
Summary
V. Image Analysis
� Part 1� Mathematical Morphology� Contour detection and analysis
� Part 2� Region-based segmentation
� Part 3� Shape analysis� Shape recognition
![Page 5: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/5.jpg)
Département GE - TI - Olivier Bernard 5
Summary
V. Image Analysis
� Part 1� Mathematical Morphology� Contour detection and analysis
� Part 2� Region-based segmentation
� Part 3� Shape analysis� Shape recognition
![Page 6: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/6.jpg)
Département GE - TI - Olivier Bernard 6
Region-based segmentation
� Context� Definition� Principle
� Two kinds of methodology� Histogram based� Region transformation based
![Page 7: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/7.jpg)
Département GE - TI - Olivier Bernard 7
� Context - Definition
Region-based segmentation
Segmentation consists on partitioning an image into a set of connected regions
Find homogeneous regions according to a specific criterion (intensity value, texture)
![Page 8: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/8.jpg)
Département GE - TI - Olivier Bernard 8
Region-based segmentation
The aim of region detection is to provide the possibility to characterize the detected object by parameter analysis (shape, position, size...)
� Context - Definition
Example of the need of region-based segmentation:
object labeling
![Page 9: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/9.jpg)
Département GE - TI - Olivier Bernard 9
Region-based segmentation
� Context - Definition
Region segmentation is an ill-posed problem
The choice of a particular method is linked to
the nature of the image (texture, intensity, ...)
the shape of the object to extract
The time constrain
the perfect segmentation does not exist
![Page 10: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/10.jpg)
Département GE - TI - Olivier Bernard 10
Region-based segmentation
Region-based Segmentation
Regiontransformation
based
- Region growing
- Region splitting
� Context – Two different methods
- Region merging
Histogram based
![Page 11: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/11.jpg)
Département GE - TI - Olivier Bernard 11
Region-based segmentation
� Context� Definition� Principle
� Two kinds of methodology� Histogram based� Region transformation based
![Page 12: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/12.jpg)
Département GE - TI - Olivier Bernard 12
Region-based segmentation
Region-based Segmentation
Regiontransformation
based
- Region growing
- Region splitting
� Context – Two different methods
- Region merging
Histogram based
![Page 13: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/13.jpg)
Département GE - TI - Olivier Bernard 13
� Histogram-based methods - Example
Region-based segmentation
The main idea is to individually select each specific histogram mode, and then select the corresponding area by thresholding
Separation of A and B classesfrom class C (background)
Initial image
A
B
C
![Page 14: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/14.jpg)
Département GE - TI - Olivier Bernard 14
Region-based segmentation
� Histogram-based methods - Example
Detection of the corresponding area by thresholding
Extraction of a specific mode of the histogram
Among the different area that contribute to this mode, select the biggest connected region from high level tools (for example: mathematical morphology)
Apply this scheme for each histogram mode
![Page 15: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/15.jpg)
Département GE - TI - Olivier Bernard 15
Region-based segmentation
� Histogram-based methods - Example
Extraction of a specific mode of the histogram
Initial image Image histogram
![Page 16: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/16.jpg)
Département GE - TI - Olivier Bernard 16
Region-based segmentation
� Histogram-based methods - Example
Extraction of a specific mode of the histogram
![Page 17: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/17.jpg)
Département GE - TI - Olivier Bernard 17
Region-based segmentation
� Histogram-based methods - Example
Detection of the corresponding area from thresholding
![Page 18: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/18.jpg)
Département GE - TI - Olivier Bernard 18
Region-based segmentation
� Histogram-based methods - Example
Selection of the biggest connected region
Proposed scheme
- Hysteresis thresholding
- Mathematical morphology(erode, fill, close)
![Page 19: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/19.jpg)
Département GE - TI - Olivier Bernard 19
Region-based segmentationCorresponding matlab code
% Read imageimg = imread(‘brain.png');img = double(img(:,:,1));
% Selection of pixels corresponding to the selected histogram modee = find(img>118);mask = zeros(size(img));mask(e) = 255;
% Region erosion through math morphse = strel('disk',3);maskErode = imerode(mask,se);
% Hysteresis thresholdingmaskHyst = hysteresis(mask,maskErode);
% Fill holes through math morphomaskFill = imfill(maskHyst,'holes');
% Region closing through math morphose = strel('disk',1);final = imclose(maskFill,se);
![Page 20: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/20.jpg)
Département GE - TI - Olivier Bernard 20
Region-based segmentationCorresponding matlab code
Selection of pixels mode
mask
% Read imageimg = imread(‘brain.png');img = double(img(:,:,1));
% Selection of pixels corresponding to the selected histogram modee = find(img>118);mask = zeros(size(img));mask(e) = 255;
% Region erosion through math morphse = strel('disk',3);maskErode = imerode(mask,se);
% Hysteresis thresholdingmaskHyst = hysteresis(mask,maskErode);
% Fill holes through math morphomaskFill = imfill(maskHyst,'holes');
% Region closing through math morphose = strel('disk',1);final = imclose(maskFill,se);
![Page 21: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/21.jpg)
Département GE - TI - Olivier Bernard 21
Region-based segmentationCorresponding matlab code
Region erosion
maskErode
% Read imageimg = imread(‘brain.png');img = double(img(:,:,1));
% Selection of pixels corresponding to the selected histogram modee = find(img>118);mask = zeros(size(img));mask(e) = 255;
% Region erosion through math morphse = strel('disk',3);maskErode = imerode(mask,se);
% Hysteresis thresholdingmaskHyst = hysteresis(mask,maskErode);
% Fill holes through math morphomaskFill = imfill(maskHyst,'holes');
% Region closing through math morphose = strel('disk',1);final = imclose(maskFill,se);
![Page 22: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/22.jpg)
Département GE - TI - Olivier Bernard 22
Region-based segmentationCorresponding matlab code
Hysteresis thresholding
maskHyst
% Read imageimg = imread(‘brain.png');img = double(img(:,:,1));
% Selection of pixels corresponding to the selected histogram modee = find(img>118);mask = zeros(size(img));mask(e) = 255;
% Region erosion through math morphse = strel('disk',3);maskErode = imerode(mask,se);
% Hysteresis thresholdingmaskHyst = hysteresis(mask,maskErode);
% Fill holes through math morphomaskFill = imfill(maskHyst,'holes');
% Region closing through math morphose = strel('disk',1);final = imclose(maskFill,se);
![Page 23: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/23.jpg)
Département GE - TI - Olivier Bernard 23
Region-based segmentationCorresponding matlab code
Fill holes
maskFill
% Read imageimg = imread(‘brain.png');img = double(img(:,:,1));
% Selection of pixels corresponding to the selected histogram modee = find(img>118);mask = zeros(size(img));mask(e) = 255;
% Region erosion through math morphse = strel('disk',3);maskErode = imerode(mask,se);
% Hysteresis thresholdingmaskHyst = hysteresis(mask,maskErode);
% Fill holes through math morphomaskFill = imfill(maskHyst,'holes');
% Region closing through math morphose = strel('disk',1);final = imclose(maskFill,se);
![Page 24: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/24.jpg)
Département GE - TI - Olivier Bernard 24
Region-based segmentationCorresponding matlab code
Region closing
final
% Read imageimg = imread(‘brain.png');img = double(img(:,:,1));
% Selection of pixels corresponding to the selected histogram modee = find(img>118);mask = zeros(size(img));mask(e) = 255;
% Region erosion through math morphse = strel('disk',3);maskErode = imerode(mask,se);
% Hysteresis thresholdingmaskHyst = hysteresis(mask,maskErode);
% Fill holes through math morphomaskFill = imfill(maskHyst,'holes');
% Region closing through math morphose = strel('disk',1);final = imclose(maskFill,se);
![Page 25: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/25.jpg)
Département GE - TI - Olivier Bernard 25
Region-based segmentation� Histogram-based methods - Example
![Page 26: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/26.jpg)
Département GE - TI - Olivier Bernard 26
Region-based segmentation
� Histogram-based methods - Example
Extraction of a specific mode of the histogram
![Page 27: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/27.jpg)
Département GE - TI - Olivier Bernard 27
Region-based segmentation
� Histogram-based methods - Example
Detection of the corresponding area from thresholding
![Page 28: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/28.jpg)
Département GE - TI - Olivier Bernard 28
Region-based segmentation
� Histogram-based methods - Example
Selection of the biggest connected region
Proposed scheme
- Hysteresis thresholding
- Mathematical morphology (erode, close)
![Page 29: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/29.jpg)
Département GE - TI - Olivier Bernard 29
Region-based segmentation
� Histogram-based methods - Example
Extraction of a specific mode of the histogram
![Page 30: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/30.jpg)
Département GE - TI - Olivier Bernard 30
Region-based segmentation
� Histogram-based methods - Example
Detection of the corresponding area from thresholding
![Page 31: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/31.jpg)
Département GE - TI - Olivier Bernard 31
Region-based segmentation
� Histogram-based methods - Example
Selection of the biggest connected region
Proposed scheme
- Hysteresis thresholding
- Mathematical morphology (erode, close)
![Page 32: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/32.jpg)
Département GE - TI - Olivier Bernard 32
Region-based segmentation
� Histogram-based methods - Example
Extraction of a specific mode of the histogram
![Page 33: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/33.jpg)
Département GE - TI - Olivier Bernard 33
Region-based segmentation
� Histogram-based methods - Example
Detection of the corresponding area from thresholding
![Page 34: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/34.jpg)
Département GE - TI - Olivier Bernard 34
Region-based segmentation
� Histogram-based methods - Example
Selection of the biggest connected region
Proposed scheme
- Mathematical morphology (erode, fill holes)
![Page 35: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/35.jpg)
Département GE - TI - Olivier Bernard 35
Region-based segmentation
� Histogram-based methods - Example
Final result
Initial image Segmented image
![Page 36: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/36.jpg)
Département GE - TI - Olivier Bernard 36
Region-based segmentation
What is the best way to choose a threshold value ?
� Histogram-based methods – choice of the threshold value
There exist a lot of methods based on probability
Example: thresholding from learning
![Page 37: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/37.jpg)
Département GE - TI - Olivier Bernard 37
Region-based segmentation
Example: thresholding from learning
Goal: Separate the two modes, by choosing the minimum histogram value between two extremities
thresh = 94
back
object
� Histogram-based methods – choice of the threshold value
![Page 38: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/38.jpg)
Département GE - TI - Olivier Bernard 38
Region-based segmentation
Model each mode through a priori distributions (for example two Gaussian models p1 et p2)
Maximization of the following energy function
1 20( ) ( ) ( )
s N
ss p n dn p n dnΓ = +∫ ∫
2 1( ) 1 ( ) ( )s F n s F n sΓ = − = + =
iFwhere is a cumulative function
� Histogram-based methods – Analytic method
![Page 39: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/39.jpg)
Département GE - TI - Olivier Bernard 39
Region-based segmentation
Maximization of function ( )sΓ
thresh = 91
Approximation of the continuous integral by a discrete version
� Histogram-based methods – Analytic method
![Page 40: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/40.jpg)
Département GE - TI - Olivier Bernard 40
Region-based segmentation
Find the threshold value which minimize the variance of the corresponding two classes
Let h[n] be a histogram function
G is the center of gravity of one class
[ ]
( )[ ]
i
i
n Ci
n C
nh n
G sh n
∈
∈
=∑
∑
var is the variance of one class
2var ( ) ( ) [ ]i
i in C
s n G h n∈
= −∑
� Histogram-based methods – Empirical method
![Page 41: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/41.jpg)
Département GE - TI - Olivier Bernard 41
Region-based segmentation
Find the threshold value that minimize the variance sum
1 2arg min(var ( ) var ( ))opts
s s s= +
After simplification, the previous problem is equivalent to maximize the following energy criteri on J(s) :
1 2
1
2 2
2
[ ] [ ]
( )[ ] [ ]
n C n C
n C n C
nh n nh n
J sh n h n
∈ ∈
∈ ∈
= +∑ ∑
∑ ∑
� Histogram-based methods – Empirical method
![Page 42: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/42.jpg)
Département GE - TI - Olivier Bernard 42
Region-based segmentation� Histogram-based methods – Empirical method
Minimization of function opts
thresh = 102
Approximation of the continuous integral by a discrete version
![Page 43: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/43.jpg)
Département GE - TI - Olivier Bernard 43
Region-based segmentation
Implementation is quite simple
Performance relatively low because of the non use of spatial coherence
Method to be used when:
� Histogram-based methods – properties
images have regions which are easily separable (objects with high contrast)
images are defined on multi-channels (multi-spectral images) which gives more information on the corresponding histogram
![Page 44: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/44.jpg)
Département GE - TI - Olivier Bernard 44
Region-based segmentation
� Context� Definition� Principle
� Two kinds of methodology� Histogram based� Region transformation based
![Page 45: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/45.jpg)
Département GE - TI - Olivier Bernard 45
Region-based segmentation
Region-based Segmentation
Regiontransformation
based
- Region growing
- Region splitting
� Context – Two different methods
- Region merging
Histogram based
![Page 46: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/46.jpg)
Département GE - TI - Olivier Bernard 46
� Region transformation based methods
Region-based segmentation
Step that consists on splitting an image into subsets Ri called regions
![Page 47: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/47.jpg)
Département GE - TI - Olivier Bernard 47
� Region transformation based methods
Region-based segmentation
Methods based on a similarity criterion P and an elementary partition
One has to define a test function that allows to validate the similarity criterion
![Page 48: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/48.jpg)
Département GE - TI - Olivier Bernard 48
� Region transformation based methods
Region-based segmentation
Similarity criterion in segmentation: region Ri is homogeneous
Test function used to verify the previous similarit y criterion
Region contrast
[ ] [ ]( ) max ( , ) min ( , )i ii R RP R true I x y I x y σ= ⇔ − <
![Page 49: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/49.jpg)
Département GE - TI - Olivier Bernard 49
� Region transformation based methods
Region-based segmentation
Test function used to verify the previous similarit y criterion
Region standard deviation
21( ) ( ( , ) )
i
iR
P R true I x y mN
σ= ⇔ − <∑
where( )iN Card R=
1( , )
iR
m I x yN
= ∑
Similarity criterion in segmentation: region Ri is homogeneous
![Page 50: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/50.jpg)
Département GE - TI - Olivier Bernard 50
� Region transformation based methods
Region-based segmentation
Test function used to verify the previous similarit y criterion
Region entropy (measure of amount of information)
( ) ( ) log( ( ))i
iR
P R true p I P I σ= ⇔ − <∑
Similarity criterion in segmentation: region Ri is homogeneous
![Page 51: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/51.jpg)
Département GE - TI - Olivier Bernard 51
� Region transformation based methods
Region-based segmentation
Methods based on a similarity criterion P and an elementary partition
A partition Π is a set of regions Ri inside an image that verify
, ( )
i j
ii
i
i j i j R R
R
i R
∀ ≠ ∩ = ∅
=
∀ ≠ ∅∪
Π ⇔ Image domain
![Page 52: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/52.jpg)
Département GE - TI - Olivier Bernard 52
Region-based segmentation
� Region transformation based methods
There exists a high number of partitioning of an image
There exists a high number of partitions that verifies the test function associated to a similari ty criterion
![Page 53: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/53.jpg)
Département GE - TI - Olivier Bernard 53
Region-based segmentation
� Region transformation based methods
How does we choose between several partitions that verify the same similarity criterion ?
The number of subsets of a partition (to minimize)
The size of the smallest sub-region (to maximize)
The distance between regions ( sum of distances between adjacent areas to maximize)
![Page 54: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/54.jpg)
Département GE - TI - Olivier Bernard 54
There exists three main groups
Region growing method
Splitting region method ( quad-tree)
� Region transformation based methods
Adjacency graph method
Region-based segmentation
![Page 55: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/55.jpg)
Département GE - TI - Olivier Bernard 55
� Region growing method
Region-based segmentation
Method based on the clustering of neighboring pixels of a region that verify a specific assumptio ns
![Page 56: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/56.jpg)
Département GE - TI - Olivier Bernard 56
Region-based segmentation� Region growing method - Principle
Initialization of a region R0 with one or more pixels (seeds)
Clustering to R0 of all the neighboring pixels which verify a specific assumption
Iteration until convergence
![Page 57: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/57.jpg)
Département GE - TI - Olivier Bernard 57
Region-based segmentation� Region growing method - Example
Example of assumption: homogeneity with the mean value of the growing region (that is updated at each iteration)
Rµ
( ) RI x thresholdµ− <
assumption
Region R0
Neighboring
( )I x
![Page 58: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/58.jpg)
Département GE - TI - Olivier Bernard 58
Region-based segmentation� Region growing method - Example
After few iterations after 5000 iterations
Convergence obtained after
10119 iterations
![Page 59: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/59.jpg)
Département GE - TI - Olivier Bernard 59
Region-based segmentation� Region growing method - Example
Mathematical Morphology
![Page 60: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/60.jpg)
Département GE - TI - Olivier Bernard 60
Region-based segmentation� Region growing method - Example
After few iterations after 5000 iterations
Convergence obtained after
10085 iterations
![Page 61: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/61.jpg)
Département GE - TI - Olivier Bernard 61
Region-based segmentation� Region growing method - Example
Mathematical Morphology
![Page 62: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/62.jpg)
Département GE - TI - Olivier Bernard 62
Region-based segmentation� Region growing method - Example
![Page 63: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/63.jpg)
Département GE - TI - Olivier Bernard 63
Region-based segmentation� Region growing method - Properties
The quality of the result directly depends on the positioning of the initial seeds
The pixel clustering order has an influence on the result
The implementation is relatively simple and the algorithm is relatively fast
![Page 64: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/64.jpg)
Département GE - TI - Olivier Bernard 64
There exists three main groups
Region growing method
Splitting region method ( quad-tree)
� Region transformation based methods
Adjacency graph method
Region-based segmentation
![Page 65: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/65.jpg)
Département GE - TI - Olivier Bernard 65
Region-based segmentation� Splitting region method
Split an image into a set of regions that verifies a specific assumption
![Page 66: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/66.jpg)
Département GE - TI - Olivier Bernard 66
Region-based segmentation
� Splitting region method - PrincipleOne starts from the whole image , called R
One applies several splitting δ which creates new sub-regions Ri
δ
For each Riδ one tests the assumption P and one
keep the best sub-region δ, that is:
the one where each sub-region verifies P
the one where there are the more sub-regions that verify P
Each sub-region that does not verify P becomes a new region R that can be process as above
![Page 67: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/67.jpg)
Département GE - TI - Olivier Bernard 67
Region-based segmentation� Splitting region method
Example of region partitioning with two partitions
Region partition: choice between horizontal or vertical partitions
Partitioned image
![Page 68: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/68.jpg)
Département GE - TI - Olivier Bernard 68
Region-based segmentation� Splitting region method - Example
Building of the corresponding Quad-treeInitial image
Quad-tree (with only square partitioning)
![Page 69: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/69.jpg)
Département GE - TI - Olivier Bernard 69
Region-based segmentation� Splitting region method - Example
Quad-tree (with only square partitioning)
Building of the corresponding Quad-treeInitial image
![Page 70: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/70.jpg)
Département GE - TI - Olivier Bernard 70
Region-based segmentation� Splitting region method - Example
Quad-tree (with only square partitioning)
Building of the corresponding Quad-treeInitial image
![Page 71: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/71.jpg)
Département GE - TI - Olivier Bernard 71
Region-based segmentation� Splitting region method - Example
Quad-tree (with only square partitioning)
Building of the corresponding Quad-treeInitial image
![Page 72: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/72.jpg)
Département GE - TI - Olivier Bernard 72
Region-based segmentation
The geometry of partitioning has a direct influence on the segmentation result
For example, the quad-tree method brings into relief square regions
There exist others partitioning strategies (triangular, pyramidal)
� Splitting region method - Properties
The choice between different partitioning strategies can be conditioned by the shape of the object to recover
![Page 73: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/73.jpg)
Département GE - TI - Olivier Bernard 73
Region-based segmentation
Initial image Building of the corresponding Quad-tree
Example: Quad-tree
Detection of the local intensity variation
![Page 74: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/74.jpg)
Département GE - TI - Olivier Bernard 74
Region-based segmentation
After the Quad-tree analysis
Building of the corresponding Quad-tree
Idea: analyze only the small blocks
Example: Quad-tree
![Page 75: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/75.jpg)
Département GE - TI - Olivier Bernard 75
There exists three main groups
Region growing method
Splitting region method ( quad-tree)
� Region transformation based methods
Adjacency graph method
Region-based segmentation
![Page 76: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/76.jpg)
Département GE - TI - Olivier Bernard 76
Region-based segmentation� Adjacency graph
Express the result obtained from a segmentation procedure into a graph structure and then exploit the graph properties in order to perform merging steps
![Page 77: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/77.jpg)
Département GE - TI - Olivier Bernard 77
Region-based segmentation� Adjacency graph - Principle
From an initial segmentation, definition of an adjacency graph where a region is a node and an arc is an adjacency relation
Definition of a similarity function between two nodes
One sorts each node couple from there energy
One merges the two best candidates
One updates the list and iterate
![Page 78: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/78.jpg)
Département GE - TI - Olivier Bernard 78
Region-based segmentation� Adjacency graph - Example
Design of a graph during a quad-tree procedure
Initial image Design of the corresponding adjacency graph
![Page 79: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/79.jpg)
Département GE - TI - Olivier Bernard 79
Region-based segmentation� Adjacency graph - Example
Initial image Design of the corresponding adjacency graph
Design of a graph during a quad-tree procedure
![Page 80: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/80.jpg)
Département GE - TI - Olivier Bernard 80
Region-based segmentation� Adjacency graph - Example
Design of a graph during a quad-tree procedure
Initial image Design of the corresponding adjacency graph
![Page 81: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/81.jpg)
Département GE - TI - Olivier Bernard 81
Region-based segmentation� Adjacency graph - Example
Design of a graph during a quad-tree procedure
Initial image Design of the corresponding adjacency graph
![Page 82: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/82.jpg)
Département GE - TI - Olivier Bernard 82
Region-based segmentation
Exploitation of an adjacency graph
Initial image Building of the corresponding region-based segmentation
� Adjacency graph - Example
![Page 83: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/83.jpg)
Département GE - TI - Olivier Bernard 83
Region-based segmentation� Adjacency graph - Example
Exploitation of an adjacency graph
Initial image Building of the corresponding region-based segmentation
![Page 84: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/84.jpg)
Département GE - TI - Olivier Bernard 84
Region-based segmentation� Adjacency graph - Example
Exploitation of an adjacency graph
Initial image Building of the corresponding region-based segmentation
![Page 85: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/85.jpg)
Département GE - TI - Olivier Bernard 85
Region-based segmentation� Adjacency graph - Example
Exploitation of an adjacency graph
Initial image Building of the corresponding region-based segmentation
![Page 86: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/86.jpg)
Département GE - TI - Olivier Bernard 86
Region-based segmentation
The order of region merging has a direct influence on the result
Generally, it is desired to merge at first the smal lest regions
� Adjacency graph - Properties
![Page 87: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/87.jpg)
Département GE - TI - Olivier Bernard 87
End of second part
Image analysis
![Page 88: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/88.jpg)
Département GE - TI - Olivier Bernard 88
ANNEXE
� Example of matlab code
![Page 89: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/89.jpg)
Département GE - TI - Olivier Bernard 89
Image analysis - Annexe
You can download all the matlab code used to generate all the presented results at the followingweb-link adress
www.creatis.insa-lyon.fr/~bernard/AnalyseImage
![Page 90: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/90.jpg)
Département GE - TI - Olivier Bernard 90
Image analysis - Annexe
List of morphological functions that can be used undermatlab 7 (type « morphology » in the documentation)
![Page 91: DIP-5 en ImageAnalysis Part2-Very Nice](https://reader034.vdocument.in/reader034/viewer/2022042519/55cf9c83550346d033aa16c9/html5/thumbnails/91.jpg)
Département GE - TI - Olivier Bernard 91
function out = hysteresis(in1,in2)
out = in2; count = 1; k=0;MAXITERATION = 200;while ( ( count ~= 0 ) && ( k < MAXITERATION ) )
count = 0;for i=2:(size(out,1)-1)
for j=2:(size(out,2)-1)if ( out(i,j) > 0 )
if ( in1(i-1,j) > 0 ) out(i-1,j) = 255; count = count + 1;
endif ( in1(i+1,j) > 0 )
out(i+1,j) = 255; count = count + 1;endif ( in1(i,j-1) > 0 )
out(i,j-1) = 255; count = count + 1;endif ( in1(i,j+1) > 0 )
out(i,j+1) = 255; count = count + 1;end
endend
endk = k + 1;
end
Matlab code of the hysteresis function (4 connexities)
Annexes – hysteresis thresholding