A new rotation invariant weber local descriptor for recognition of skin diseases

Abstract

A new rotation invariant Weber Local texture Descriptor(WLD) is proposed here. Performance of the developed features is evaluated on the basis of the recognition accuracies of SVM classifiers, trained and tested with the extracted features from the images of skins affected with three popular skin diseases such as Leprosy, Tineaversicolor, Vitiligo, and also normal skin, collected from School of Tropical Medicine, Kolkata. The WLD features are extracted with variations of the radius from 1 to 3 considering perimeters of having 8, 16 and 24 pixels respectively. The modified WLD provides an average improvement of 4.79% in recognition accuracy over the normal WLD in the present four class problem. Dividing each sample image into 4 sub-regions through its centre of gravity and extracting WLD features from each of them. Thus we have extracted two different feature sets having normal WLD and rotation invariant WLD. They provide the maximum recognition accuracies of 85.06% and 87.36% respectively using SVM classiifiers on test set. © Springer-Verlag 2013.