Texture Classification Based on Cooccurrence Matrices Presentation III

Texture Classification Based on Co-occurrence Matrices Presentation III Pattern Recognition Mohammed Jirari Spring 2003

Co-occurrence Matrices n n n The joint probability of occurrence of grey level a and b for two pixels with a defined spatial relationship in an image. The spatial relationship is defined in terms of distance d angle θ. From these matrices, a variety of features may be extracted.

Co-occurrence Matrices (cont. ) n n In my project, the matrices are constructed at a distance of d=1 and for angles θ=0°, 45°, 90°, 135°. For each matrix, eight features are extracted.

Co-occurrence Matrices (cont. ) n Can be formally represented as follows:

Example n A 4 X 4 image with 4 grey-levels

Features Used n Energy or angular second moment: n Entropy: n Maximum Probability: Inverse Difference moment: κ=2, λ=1 n

Features Used (cont. ) n Contrast: n Homogeneity: n Inertia or variance:

Features Used (cont. ) n Correlation

Matlab Code to extract features from images Co-occurrence and features

Results n n n n Features masses Features Features for Calcification for Well-defined/Circumscribed for for for Spiculated masses other, ill-defined masses Architectural distortion Asymmetry Normal mammogram

What next? n n I plan to use the calculated features as training sets for my neural network, reducing the training set size from 1024 X 1024 to 8 X 4 per image. Also, a fifth co-occurrence matrix will be constructed as the mean of all four directions. May or may not help !!!