Texture information fusion based image classification

Abstract

The classical gray level co-occurrence matrix(GLCM) neglects the directional differences of texture. A novel method of image classification is presented in this paper. Multi-angle weighting of GLCM is proposed to decrease effects of directional difference, in which second-order statistics, such as energy, entropy, moment of inertia, moments of deficit and relevance, are calculated to describe image texture. The contributions to classifier of second-order statistics are evaluated using information gain. The model based on one against one support vector machines (OAOSVM) is realized. The experimental results has shown that the proposed method can accomplish image classification with higher accuracy on several standard image datasets than other methods.