A Fast Search Method for Vector Quantization Using Enhanced Sum Pyramid Data Structure

摘要

Conventional vector quantization (VQ) encoding method by full search (FS) is very heavy computationally but it can reach the best PSNR. In order to speed up the encoding process, many fast search methods have been developed. Base on the concept of multi-resolutions, the FS equivalent fast search methods using mean-type pyramid data structure have been proposed already in 2-4. In this Letter, an enhanced sum pyramid data structure is suggested to improve search efficiency further, which benefits from (1) exact computing in integer form, (2) one more 2-dimensional new resoludon and (3) an optimal pair selecting way for constructing the new resolution. Experimental results show that a lot of codewords can be rejected efficiently by using this added new resolution that features lower dimensions and earlier difference check order.