Iterative collage coding for fractal compression

A fractal encoder processes the original source vector by selecting a contractive map whose unique fixed point (attractor) approximates the original. A description of the map is transmitted to the decoder. The decoder iterates the map to recover the attractor. The task of picking the map is called the inverse problem. The predominantly used, suboptimum, technique to solve the inverse problem is collage coding. In collage coding, only the first decoding iteration is optimized. We propose two new suboptimum algorithms for the inverse problem. In our approach, the encoder imitates the iterative operation of the decoder. At each step, however, a new map is used so as to keep the sequence of approximations close to the original. If the sequence converges, then the limit is an attractor, and it is a good candidate for the reconstructed vector. The description of the map corresponding to the attractor is sent to the decoder