Improving the performance of hierarchical vector quantization using segmentation

Hierarchical vector quantization (HVQ) is a constrained VQ technique where both the encoder and decoder are implemented using only table-lookups. HVQ is extremely fast as well as easy to implement in both software and hardware. But the rate-distortion performance of HVQ is poor. In this paper, we combine rate-distortion optimal binary tree segmentation with the HVQ algorithm. An algorithm for segmenting a block entirely through table-lookups is presented. Simulation results indicate that this algorithm can gain up to 3.5 dB over plain HVQ.