一种新的矢量量化码书设计算法  

A New VQ Codebook Design Algorithm

作　　者：黄榜[1] 谢林柏 

机构地区：[1]江南大学轻工过程先进控制教育部重点实验室,系统工程研究所,无锡214122

出　　处：《科学技术与工程》2011年第1期46-49,55,共5页Science Technology and Engineering

基　　金：国家自然科学基金(60804013);中央高校基本科研业务费专项资金(JUSRP21011)资助

摘　　要：提出了一种针对码书优化的图像矢量量化算法。首先设定矢量距离的初始门限值,基于空间划分对训练矢量聚类,找到矢量个数小于平均胞腔矢量数的胞腔;提取其聚类矢量后删除该胞腔,然后缩小距离门限值再次聚类提取,依次循环直到聚类矢量个数达到要求。将提取的聚类矢量作为初始码书,进行LBG算法码书设计,有效地改善了传统LBG算法依赖于初始码书而容易陷入局部最优的缺点,采用改进算法生成的码书更接近全局最优,加快了收敛速度。仿真结果证明了该方法的有效性。An optimized VQ codebook design algorithm of image compression is introduced in this paper.First of all,set the initial distance threshold,and cluster the train vectors base on space partition,collect the clustery vectors which has more train vectors in its cell lumens than average vector number,and delete these cell lumens.Secondly,lessen the distance threshold and cluster again to obtain more clustery vectors until the number of clustery vectors meet requirement.Finally,consider these clustery vectors as the initial codebook for LBG algorithm. Though the efficiency of traditional LBG algorithm depends on the selecting of initial codebook,we can still obtain a codebook which close to global optimal and have a faster rate of convergence after we improved the initial codebook.The validity is proved by the simulation result.

关 键 词：矢量量化 初始码书 LBG算法 聚类算法 

分 类 号：TN919.81[电子电信—通信与信息系统]