Information-theoretic approach to joint separation and segmentation of images
Hichem Snoussi · 2006
This contribution is devoted to the problem of blind joint separation and segmentation of images in linear mixtures. One key point concerning source separation is the exploitation of the non stationarity of the sources. In the case of 2-D signals (images), this is based on a joint separation and segmentation of sources. One of the objectives of this work is to propose a robust solution (a learning machine) based on the maximization of mutual information and higher order statistics in order to incorporate the segmentation in a information theoretic approach. The joint separation and segmentation has a double goal: (i) the separation can be considered as an extension of the segmentation in the situation of mixed sources and (ii )t he segmentation can be considered as an efficient tool to exploit the non stationarity in a separation perspective.