一种改进的等误差竞争学习矢量量化算法  

An improved competitive learning vector quantization algorithm based on equidistortion

作　　者：徐皓淋[1] 陈善学[1] 

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065

出　　处：《重庆邮电大学学报（自然科学版）》2009年第6期721-724,753,共5页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：国家自然科学基金(60602057);重庆市教委基金项目(KJ070511)

摘　　要：针对等误差竞争学习矢量量化算法的初始码书生成随机性较强和搜索获胜码字计算量较大这2个缺点,提出了一种改进算法。对于初始码书的缺点,改进算法采用一种基于训练矢量的统计特征量的分类平均初始码书生成算法,同时改进算法利用3个不等式来快速排除大量候选码字,从而解决了原算法计算量较大的问题。仿真实验表明,改进算法的计算量比原算法减小了80%,而图像效果即峰值信噪比(PSNR)比原算法平均提升了0.5 dB左右。Aiming at the problems of high randomicity for the formation of initial eodebook and large computational complexity for searching codeword in competitive learning vector quantization algorithm based on equidistortion, an improved algorithm was proposed. For the defects of initial eodebook, the improved algorithm adopts the grouping codeword formation method based on the statistical features of training vectors, and uses three inequalities to eliminate the unnecessary code-word. Simulation shows that the reduction of computation is substantial and the codebook performance is improved. Compared with the former algorithm, the computational complexity reduces by 80% and the encoding quality can be improved by 0.50dB.

关 键 词：矢量量化 等误差原则 初始码书 快速搜索 

分 类 号：TN911.21[电子电信—通信与信息系统]