基于特征值排序的部分失真快速码字搜索算法

Fast Codeword Search Based on Ranking of Eigenvalue

提出了一种矢量量化快速码字搜索算法。该算法在编码前预先计算每个码字的特征值并按顺序排列;在编码时,根据每个输入矢量的特征值来确定码字搜索顺序。同时限定相应的搜索范围及利用有效的码字删除准则,从而大大提高了编码速度。实验表明,该算法只需要穷尽算法2%-4%的编码时间就可以获得与之较为接近的编码质量,编码速度与ASRSS算法及MEENNS算法相比也有明显提高。

A fast codeword search algorithm for vector quantization is presented in this paper. Before coding, the eigenvalue of each codeword is computed and ranked; during coding, the search sequence is decided by the eigenvalue of each input vector. Meanwhile the search range is confined and efficient elimination criterion is used, thus to greatly accelerate the coding process. Simulation results show that this algorithm achieves a performance close to that of full search while the coding time is only 3%-4% of the full search process. Also, the coding process is largely accelerated as compared with that of ASRSS algorithm and MEENNS algorithm.