摘要
经典LBG算法的局部极小值问题是制约其性能的重要因素.根据渐进最优矢量量化理论[4]的思想提出了一种改进型LBG算法,它采用码字转移的方法使各个类的畸变趋于平衡,从而近一步减小平均畸变以获得性能更优的量化器.文中介绍了若干实验,对多种分布的样本以及2维图像进行了经典算法和改进型算法的比较.从实验结果看出,后者的算法性能大大优于前者.
The local minimum problem is a much important factor which restricts the performance of the classic LBG algorithm, This article puts forward an improved LBG algorithm on the basis of the asymptotically optimal vector quantization theory. To obtain a quantizer with higher quality, the algorithm uses the method of codewords shifting which can ulteriorly diminish the average distortion. The article also introduces several experiments using samples with different probability distribution and 2D images to compare the performance between the classic and improved algorithm. It is obvious from the results that the latter is better.
出处
《小型微型计算机系统》
CSCD
北大核心
2005年第9期1566-1570,共5页
Journal of Chinese Computer Systems
