Design and analysis of predictive reflected residual vector quantization

Image communication is primarily constrained by its large bandwidth requirement. Therefore, researchers have worked on various compression algorithms to achieve low bit rates. Images and video sequences are highly-correlated sources and their correlation should be exploited in a given compression algorithm. Differential pulse code modulation (DPCM) emerged as a means of exploiting the correlation among image pixels. Later, DPCM was improved by predictive vector quantization (PVQ). PVQ employs block by block prediction and results in satisfactory performance at low bit rates. However, its design is complicated and recently an asymptotic closed-loop (ACL) was proposed to stabilize the design. We attempt to replace the VQ with a multistage VQ structure in the hope of further reducing the stress on the closed-loop design. We use the multistage VQ structure called reflected residual vector quantization (RRVQ). RRVQ works by imposing an additional symmetry constraint on the multistage codebook design. RRVQ has been quite popular where large block-length vector quantization is needed due to its very low codebook search capability. Our proposed design goal in replacing VQ with RRVQ in a PVQ design is our wish to use large block lengths like 16/spl times/16 or 32/spl times/32 size vectors to grab any linear/nonlinear correlation among the vector components. The way to incorporate RRVQ within PVQ structure is proposed and simulation results are discussed.