Texture Classification Based on Topographic Image Structure

In this paper, we present a structural and statistical approach for texture classification. It can achieve with a higher accuracy rate comparing to the Spatial Gray Level Dependence (SGLD) method and Laws' method. Our method uses the previously proposed operator, the Surface-Shape operator (SS-operator), for describing topographic structure of texture images. The SS-operator describes shape of each pixel comparing with its neighbourhood in terms of topographical shapes such as hill, dale, ridge, valley, etc. Then, we use co-occurrence matrices, a statistical measure, to summarize the statistical distributions of such topographical shapes over the considered image to form texture features. This method yields good classifications of MIT vision textures.