Multi-model Approach for Multicomponent Texture Classification

This paper concerns multicomponent texture classification. The aim is to provide a flexible model when wavelet subband coefficients of components do not have the same distributions. Example of such case is when color textures are represented in a perceptual color space. In this kind of representation, the separability between luminance and chrominance components have to be considered in the modeling process. The contribution of this work consists in proposing a multi-model based characterization for this type of multicomponent images. For this, two models ML and MCr are used in order to extract features from luminance and chrominance components, respectively. We discuss in detail and define the multi-model when textures are represented in the HSV color space as a special case of multicomponent analysis. Experimental results show that the proposed approach improves performances of the classification system when compared with existing methods.