Exploring information theory and gaussian markov random fields for color texture classification

Abstract

This paper proposes a novel approach to compute information theory measures from Gaussian Markov Random Field (GMRF) for color texture classification task. We firstly transform the three color channels of the input image into three set of sub-bands of the form LL, HH, HL and LH using three Discret Wavelet Transforms. We then visualize each sub-band as a GMRF from which we generate features by computing Fisher information matrix and Shannon’s entropy to encode the local spatial dependency. The concatenation of the computed features are then used as the texture descriptor, which in turn is used as input for the classifiers referred to in this work. Experiments were performed with color texture images from public databases widely used in the literature that demonstrate the efficiency of the proposed method.