Abstract: In this paper,A method of vector quantization with adaptive difference compensation based on dynamic image block regularization is presented. The method analyzes the similarity of the encoding image block and its eight-neighbour blocks, then decides the encoding method according to the comparison of the similarity value with the known threshold . If the similarity value is more than the threshold,the neighbour block and the encoding image block uses the same codeword to encode, otherwise , the neighbour block singly encodes. When encoding, the method counts the difference between the image block and the matching codeword, obtains the difference image, then gets the pixel difference sign bits, carries on the running length coding and attaches them after the codeword index; When decoding,the method restores the image according to the codeword index,has the running length encoding and carries on the adaptive difference compensation with the window of size, obtains the final image.Finally, the experiment results show that the method in this paper can improve the encoding speed and image restoration performance against the normal vector quantization.

Key words: Similarity, Eight-Neighbour, Codebook, LBG Algorithm, Vector Quantization

摘要： 提出一种图像块动态调整的自适应差值补偿矢量量化方法。该方法在编码前，分析待编码子块与其八邻域子块的相似度，通过事先给定的阈值判断子块是否相似，若相似，则用相同码字编码；否则单独编码。在编码时，计算子块和匹配码字的差值，得到差值图像，进而得到像素差值符号位，对其进行行程编码后附在码字索引之后；在解码时，根据码字索引恢复图像，对附加信息进行行程解码，并采用 补偿窗口对每个恢复像素进行自适应差值补偿，得到最终恢复图像。实验结果表明，相对于普通矢量量化，论文方法不但可以提高编码速度，而且图像质量有明显改善。

关键词: 相似度, 八邻域, 码本, LBG算法, 矢量量化