A Partial Search Mechanism for VQ-based Codebook Match

Abstract

In this paper, a mechanism for accelerating the search on VQ-based codebook match is proposed. With our approach, all pixel blocks of vector representation in an image picture could be encoded efficiently into their corresponding indices, respectively, associated the closest codeword in the pre-generated codebook. The technique adopted in our scheme is inspired from the concept of space partition to the initial codebook, in the manner that the search range for the image block is reduced significantly. There is a key-codebook generated from the given codebook in system initialization comprised of numerous key- codewords, its size is smaller than the original codebook. In our scheme, the input image block is first directed to look for the closest key-codeword in the key-codebook. After that, a dominated set associated with the closest key-codeword is fixed. The purposed codeword closest corresponding to the input image block in the specified dominated set is ultimately picked up. In such a way, the time to obtain the exactly closest codeword is then shortened when comparing the traditional full VQ codeword search. Moreover, we propose a skill of spreading the dominated area in order to attain the most precise hit radio estimation for the closest codeword. Accordingly, the advantages we achieve are at least two and half times faster than that of full search in VQ implementation, as observed from the experiments. Besides, the mechanism we propose is also compatible with the search algorithms explored in past relevant literature. Thereby, our proposed scheme is feasible and extendable in VQ codeword match.