Image Classification Schemes Based on Gradient Matrix and Contrast Matrices

Abstract

Texture classification and categorizing are used in various pattern recognition applications and classification texture that possesses a characteristic appearance. This work aims to provide an improved scheme of enhanced classification decision with the need to increase the precision time significantly. This research studied the discriminating characteristics of textures by extracting the feature from gradient matrix (GM), the features were extracted using the first-order gradient feature vector, three Gradient Matrices were established, one for Max value, another for Min value and last was the Average value, these matrices were calculated by extracting the gradient along x-axis and y-axis and the gradient along the diagonal. A feature vector consist of 210 features was calculated to represent each image sample and contrast matrix CM, The feature extracted from CM1 was The difference between the sum of the neighborhood values of 3x3 pixels those larger than the pixel values (center pixels) divided by their number and the sum of the neighborhoods values of 3x3 pixels those smaller than the pixel value divided by their number total feature vector was 210, Four types of Euclidean distance metrics were used for classification decision purposes. The concepts “average” and “standard deviation” were calculated to perform the interentra scatter analysis for each feature to find out the best discriminating features that can be used. The final result of the test set of GM is 98.3 while training set was 97.3, the final result of the test set of CM is 98.2 while training set was 95.7.