Trellis-coded residual vector quantization

Type Thesis

Rate 0.0 /5

Details

Files

Reviews

Download

Summary

Residual vector quantization (RVQ) is investigated in context of quantization cell shapes and is found to produce oblong cell shapes. The presence of these oblong cell shapes are in part responsible for the degradation in performance of RVQ compared to vector quantization (VQ). In an attempt to provide better cell shapes for the RVQ a trellis-based design is suggested in this thesis, referred to as trellis-into-trellis-coded RVQ. Trellis-into-trellis design provided better simulation results than the sequential design but had a large computation complexity. In order to alleviate the large computation complexity, trellis-coded residual vector quantization (TCRVQ) is introduced. The new method employs direct sum partitioning and a single trellis structure. TCRVQ provides improved performance over the RVQ due to its ability to realize more spherical like quantization cells. The performance edge of TCRVQ over the sequential design stems from the use of multipath searching and the joint optimization. Furthermore, TCRVQ is combined with an entropy coder to form entropy-constrained TCRVQ. The application of entropy-constrained TCRVQ to images and subband coding are also presented.