AUTOMATED DETECTION OF SKIN DISEASES USING TEXTURE FEATURES
Anal Kumar Mittra, Ranjan Parekh · 2011
This paper proposes an automated system for recognizing disease conditions of human skin in context to health informatics. The disease conditions are recognized by analyzing skin texture images using a set of normalized symmetrical Grey Level Co-occurrence Matrices (GLCM). GLCM defines the probability of grey level i occurring in the neighborhood of another grey level j at a distance d in direction θ. Directional GLCMs are computed along four directions: horizontal (θ = 0o), vertical (θ = 90o), right diagonal (θ = 45o) and left diagonal (θ= 135o), and a set of features computed from each, are averaged to provide an estimation of the texture class. The system is tested using 180 images pertaining to three dermatological skin conditions viz. Dermatitis, Eczema, Urticaria. An accuracy of 96.6% is obtained using a multilayer perceptron (MLP) as a classifier.