中文摘要：

传统的矢量量化编码方法总是将待编码矢量以码书中唯一的最匹配码字作为其近似输出矢量,以实现数据压缩的目的.这种方法对远离码字的矢量无法避免显著的误差.本文提出组合编码的矢量量化方法,其思想是对远离码字的矢量进行主辅组合编码,对主码字编码造成的误差通过辅码字加以补偿.实验表明,该方法在很小降低压缩比率的条件下显著提高了矢量编码精度,能够在信号处理等领域发挥有效作用.

英文摘要：

Vector quantization is an important technology in the field of information and coding theory.Traditionally,all the vector quantization algorithms encode a vector by assigning just one optimal matching codeword as its representative to attain the objective of data compressing.However,for vectors lay far away from the code words,this strategy will introduce significant error inevitably because of the Intrinsic shortage of code words in the codebook.In depth encoding experiments using LBG algorithm on standard test images we have conducted showed that remarkable errors were introduced at the encoding stage,which was unrecoverable after image encoding finished.This paper put forward a combinational encoding algorithm which employs main and adjuvant code words to encode such kind of vectors.At the encoding stage,the algorithm first finds the optimal code word for the vector and,meanwhile,computes the mean square error introduced by the vector quantization algorithm.By setting threshold for the error,the scheme filters out vectors that have mean square errors large than the admitted threshold.For such kind of vectors,a scheme of error adjusting is proposed.In the scheme,all code words along with the residual vector are normalized onto the unit sphere and the most similar unit vector can be found.By finding the most similar unit vector,the prospective adjuvant codeword can be determined.With the prospective adjuvant codeword in hand,error introduced by the optimal main codeword can be adjusted by a factor of such prospective adjuvant codeword.Scheme on how to find prospective adjuvant codeword and value of factor is discussed in detail in the paper.Sets of experiments show that our method can notably improve the accuracy of vector encoding with a small portion lose of the rate of compressing.The purposed encoding strategy is worthy for further research in such fields as signal processing and image compressing.