A combinational encoding algorithm for vector quantization
Hu Yun,Xie Jun-Yuan,Wang Chong-Jun
DOI: https://doi.org/10.13232/j.cnki.jnju.2011.05.008
2011-01-01
Abstract: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.