Kullback-Leibler distance between complex generalized Gaussian distributions

In texture classification, feature extraction can be made in a transform domain. A possibility to preserve the translation invariance is to use a complex transform like the Hyperanalytic Wavelet transform. It exhibits a circularly symmetric density function for subband coefficients so it can be modeled by a particular form of the complex generalized Gaussian (CGGD) distribution function. The Kullback-Leibler (KL) divergence, or distance, can be used to measure the similarity between subbands density function. We derive in this paper a closed-form expression for the KL divergence between two complex generalized Gaussian distributions.