A hybrid statistical approach for texture images classification based on scale invariant features and mixture gamma distribution

Abstract

Image classification refers to an important process in computer vision. The purpose of this paper is to propose a novel approach named GGD-GMM and based on statistical modeling in the wavelet domain to describe textured images and rely on a number of principles that give its internal coherence and originality. Firstly, we propose arobust algorithm based on the combination of the wavelet transform and Scale Invariant Feature Transform. Secondly, we implement the aforementioned algorithm and fit the result using the finite mixture gamma distribution (GMM). The results, obtained for two benchmark datasets show that the proposed algorithm has a good relevance as it provides higher classification accuracy than some other well-known models (Kohavi, 1995). Moreover, it shows other advantages relied upon Noise-resistant and rotation invariant.