Color texture characterization based on the extended relative phase in the complex wavelets domain

In this paper, we present a color texture characterization by studying novel information of complex wavelet coefficients which we call extended relative phase (ERP) information. By definition, this latter permits to consider relation between color subbands by using straightforward univariate models, and thus avoiding cumbersomeness of multivariate fitting. The ERP information is modeled by three well known circular distributions namely, VonMises (VM), Wrapped Cauchy (WC) and Vonn. All the three models present, advantages of simple feature extraction (using maximum likelihood estimation) and similarity measurement (known forms of Kullback-leibler divergence) steps, which extremely helps for the runtime issue. Experimental results on the Vistex database show that considering ERP inter color subbands improves retrieval performances over the use of conventional relative phase intra grayscale subbands.