A new multilevel codebook searching algorithm for vector quantization

Abstract

A new multilevel codebook searching (MCS) algorithm for vector quantization is presented. Although it belongs to the category of the fast nearest neighbor searching (FNNS) algorithms for vector quantization, the new MCS algorithm is not a variation of any existing FNNS algorithms (such as k-d tree searching algorithm, partial distance searching algorithm, triangle inequality searching algorithm...). The searching strategy involves several search levels. Each level stores a certain size codebook. Searching starts from the stage containing the smallest size (lowest bitrate) codebook to the level containing largest size (highest bitrate) codebook. The searching paths between any two adjacent levels are built by using training sets. The simulation result of applying MCS algorithm to image VQ shows that the MCS algorithm can reduce searching complexity to less than 3% of an exhaustive searching VQ (ESVQ) (codebook size of 4096) while introducing negligible error (0.064 db degradation from ESVQ). A comparison between the MCS algorithm and several k-d binary tree searching algorithms is presented too. The MCS algorithm fits very well into multilevel codebook VQ in the vector transform and vector sub-band domains.