Abstract: On the basis of high coding quality，reducing encoding time and cutting down bit rates are the important problems of current research in vector quantization.Fast codeword search algorithm is an important technology to reduce encoding time.This paper presents a Modified Hadamard-Transform based Equal-average Equal-variance Nearest Neighbor Search algorithm（MHTEENNS）.The experiment shows that this algorithm can exclude more codeword and is more efficient.In order to reduce bit rates，and reduce more encoding time，fast correlation vector quantization algorithm is presented，but coding quality is declined.This paper presents a Modified Diagonal based Fast Correlation Vector Quantization algorithm（MDFCVQ）.The experiment shows that the algorithm can improve the quality by 0.8~0.9 dB and reduce bit rates.Finally，this paper combines fast codeword search algorithm and correlation vector quantization.Compared with previous algorithm，a new algorithm is presented，which has higher quality and lower bit rates.

Key words: Vector Quantization（VQ）, codeword search, fast correlation, correlation predictive, image coding

摘要： 在矢量量化中，保证编码质量的前提下，缩短编码时间和降低码率是当前研究的重要问题。快速码字搜索算法是减少编码时间的重要技术。提出了一种改进的哈达玛变换域等均值等方差最近邻搜索算法（MHTEENNS）。测试结果表明，这种算法能够排除更多的码字，效率更高。为了降低码率和进一步缩短编码时间，目前已有相关矢量量化的图像编码算法，但是这种算法造成编码质量的下降。提出了改进的基于对角线相关矢量量化编码算法（MDFCVQ）。该算法编码质量提高了0.8~0.9 dB且码率进一步降低。最后，将快速码字搜索算法应用到相关矢量量化中来，将两种改进后的技术结合在一起，通过与之前的方法比较，提出一种在保证编码时间的前提下，具有更高编码质量和更低码率的矢量量化算法。

关键词: 矢量量化, 码字搜索, 快速相关, 相关预测, 图像编码