Image coding using entropy-constrained residual vector quantization

The residual vector quantization (RVQ) structure is exploited to produce a variable length codeword RVQ. Necessary conditions for the optimality of this RVQ are presented, and a new entropy-constrained RVQ (ECRVQ) design algorithm is shown to be very effective in designing RVQ codebooks over a wide range of bit rates and vector sizes. The new EC-RVQ has several important advantages. It can outperform entropy-constrained VQ (ECVQ) in terms of peak signal-to-noise ratio (PSNR), memory, and computation requirements. It can also be used to design high rate codebooks and codebooks with relatively large vector sizes. Experimental results indicate that when the new EC-RVQ is applied to image coding, very high quality is achieved at relatively low bit rates.