Research Of Vector Quantization On Image Compression

Vector quantization techniques on image compression were studied in the thesis. The vector quantization's theory, conception, key techniques and present research were introduced at first. Then two key techniques of vector quantization--codebook design algorithms and codeword searching algorithms were mainly studied. Furthermore the thesis discussed techniques on overcoming block artifacts of reconstructed image, and gave some improved algorithms.In the part of codebook design algorithms, several codebook design algorithms were introduced firstly, And the k-means codebook design algorithm and the codebook design algorithm based on self-organizing feature map neural network were analyzed and compared following. Also, the thesis modified the updata parameter based on two improved k-means codebook design algorithms, and applied inequation critetions and up-and-down search algorithm in codebook design algorithm. The simulation results showed that the proposed algorithm in this thesis was good.In the part of codeword searching algorithms, the thesis discussed fast codeword search algorithms based on absolute error inequality, triangle inequality, norm inequality and especially the algorithm based on equal-average nearest neighbor search and equal-average equal-variance nearest search. Based on those fast codeword algorithms, the thesis decreased the range of search codeword using triangle inequality and equal-average equal-variance nearest search, and proposed a new mode of computing character parameter based on the half number of vector pixels to decreased the computation.At last, reasons of block artifacts in restructed image using vector quantization were analyzed and a new difference compensation algorithm based on vector quantization was proposed by studying two traditional improved algorithms.