Improvement of region-based image coding by neural networks

Region-based image coding is a very promising technique that is gaining considerable interest. One of the criticisms often associated to this class of image compression algorithms is its lack of realism in what concerns textures. We study an approach intended to attack this problem, relying on the encoding of the residual image (obtained after region-based encoding) using neural networks. Although preliminary, the results that we obtained, part of them presented in this paper; show that the proposed association of the two techniques is promising. Moreover due to its cascaded operation, incremental transmission arises naturally