An iterative technique for universal lossy compression of individual sequences

Universal lossy compression of a data sequence can be obtained by fitting to the source sequence a "simple" reconstruction sequence that can be encoded efficiently and yet be within a tolerable distortion from the given source sequence. We develop iterative algorithms to find such a reconstruction sequence, for a given source sequence, using different criteria of simplicity for the reconstruction sequence. As a result we obtain a practical universal lossy compression method. The proposed method can be applied to source sequences defined over finite or continuous alphabets. We discuss the relation between our method and quantization techniques like entropy coded vector quantization (ECVQ) and trellis coded quantization (TCQ).