| Summary: | 碩士 === 大同大學 === 通訊工程研究所 === 95 === Side-match vector quantization is a well-known technique for image compression. It decreases the bit rates but still has an acceptable image quality [2]. It duplicates the upper and left side boundary pixels of the coding block from the pre-coded blocks to create a sub-codebook which is much smaller than the super codebook. However, most of times, the duplicated pixels are not precisely enough for the coding block so that the sub-codebook also doesn’t adapt for the coding block.
A novel vector quantization technique is proposed in this paper. In order to not only decrease the bit rates but also create an adaptable sub-codebook for the coding block, we use the lapped vectors of the pre-code blocks and code words to extract several nearest codewords from the super codebook to create a sub-codebook for each coding block. In this way, the sub-codebook will more adapt to the coding block.
In the experimental, we use gray still images which the sizes are 512 by 512 pixels with the super codebooks sized 256 and 512 code vectors. The experimental results show that the proposed scheme has a low bit rates and better image quality than side-match vector quantization.
|
|---|
