新型混合矢量码书快速匹配算法

An New Hybrid Method of Fast Vector Quantization

矢量量化码书快速匹配算法是矢量量化技术实时应用的关键所在。从原理上看,提高矢量码书匹配速度的方法可以分为两类:一类是通过建立某种“剔除”条件来避免不必要的输入矢量和码书矢量之间欧氏距离计算“软”算法,另外一类是通过查找表(LUT)技术来代替欧氏距离计算中乘法运算的“硬”方法。文章利用这两类方法的互补性,提出了一种“软”、“硬”方法相结合的矢量码书快速匹配算法。实验表明,该算法具有非常好的实时性能,能够方便地应用于加快数据压缩矢量量化的码书匹配过程以及模式识别中SOM型神经网络的识别过程。

Fast nearest-code searching method is essential for the real-time application of the technology of Vector Quantiza-tion(VQ).The methods used to speed up the searching process of nearest-code can be categorized into two,one includes those which avoid the unnecessary computation of Square Euclidean Distance between the input vector and the codeword vectors,this category can be called a kind of“Soft”method;the other in-cludes those which use the computation Look-up Table to re-place the multiplication operations in the calculation of Square Euclidean Distance thus save much computation time,it can be called a kind of“Hard”method.Note that these two types of methods are complementary,thus a new hybrid method which combines the“Soft”method and“Hard”method is proposed.The experiment results verified the efficiency of this method.This method can be used in real-time signal processing system based on VQ.Also,it can be used in speed up those pattern recognition systems which use SOM such as a Network Security System which uses SOM to detect the possible attack of hacker.