Multi-Band Texture Modeling Using Mixture of Multivariate Generalized Gaussians and Applications

In this paper, we present a unified statistical model for multivariate and multi-modal sub-band distribution in multi-resolution transforms of color-texture images. This model is based on the formalism of finite mixtures of multivariate generalized Gaussians (MoMGG) which enables accurate representation of joint statistics of different sub-bands. The MoMGG not only enables to express correlation between sub-bands at different scales and orientations, but also between adjacent locations within the same sub-bands to better describe the spatial texture layout. Moreover, it enables to combine different multi-scale transforms (e.g., contourlets,wavelets) to build a richer and more representative texture signature. We successfully applied our model to color-texture image retrieval and reconstruction where promising results have been obtained comparatively to state-of-art methods