Method for texture classification using image structural features

In this paper, novel method for texture analysis and classification based on using image structural properties is proposed. It is known that the Human Visual System successfully perceives an image content by visualized collection of existing edges. In this paper, we propose to use the collection of gradient magnitudes as the characteristics of the structure of an image. Gradient magnitude is assumed to be a random variable with two-parameter Weibull distribution, and as a characteristic of the proximity of the two images we use a special measure of proximity of the parameter estimates of the corresponding distributions. Classification is made by method of comparison with the etalon using the database of the University of Heidelberg (Germany) which contains 10 texture classes. As an etalon is used the set of average values of the parameters estimates of the Weibull distribution, calculated by the magnitude gradients of training set of texture, and comparison is performed using the proposed proximity measure. The analysis revealed two virtually indistinguishable classes, and for the remaining eight classes it is shown that the classification error on the average is about 18%.