Entropy-constrained residual vector quantization

Abstract

The authors introduce a novel variable-rate residual vector quantizer (RVQ) where the source entropy is exploited. Necessary conditions for the optimality of this RVQ are presented as well as a novel entropy-constrained RVQ (EC-RVQ) design algorithm EC-RVQ is shown simultaneously to outperform entropy-constrained VQ (EC-VQ) in terms of performance, memory, and computation. The low cost of EC-RVQ permits the use of high-rate codebooks and/or codebooks with relatively large vector sizes (such as 8*8).< >