Abstract: A lot of correlations exist in original codebook by rotating each codeword to different directions，so a theory that an original codebook can be compressed by rotating codewords to different directions in Vector Quantization（VQ） algorithm is proposed.In this theory，all codewords in codebook are rotated to four hieratical directions to check their correlations.If correlations between two codewords with four directions exist，one codeword will be deleted from this codebook.So in compressed codebook，each codeword has four directions，it can be instead of four codewords.While encoding，the compressed codebook will be recovered by rotating codewords.An algorithm how to compress codebook by rotating codewords is also proposed and verified in this paper，a codebook can be compressed from 1 024×16×8 bit to 256×16×8 bit using this algorithm.Simulation results show that this codebook can decrease 75% memory room and I/O bandwidth but only lost 0.28 dB PSNR compared with the original codebook.

Key words: vector quantization, codebook, rotating compression

摘要： 普通码书中的码字之间在不同的方向上具有很大的相关性，存在大量的数据冗余。提出了将码书中的码字旋转压缩的理论。该理论是将各个码字按四个方向垂直旋转后进行相似性检查。如果旋转后的码字其中一个方向上与前面的码字存在相似，则将该码字删除，从而达到压缩的目的。编码时将压缩后的码书旋转恢复后进行编码，从而大幅降低了需要存储的码字数量。同时给出了一种将现有1 024阶16维码书旋转压缩成256阶16维的方法，并对该方法得到的码书性能进行了仿真验证。实验结果表明使用压缩后的码书在硬件实现时与普通的矢量量化码书相比减少了75%的存储空间和输入带宽，而PSNR平均只降低0.28 dB。

关键词: 矢量量化, 码书, 旋转压缩