Design of Vector Quantization Code Book Based on Clone Selection Algorithm and its Application

Abstract

In order to improve the drawback that the LBG algorithm is more sensitive to the initial code book, the similarity measure of Euclidean distance only reflects the local consistency characteristics of the clustering results, but not the global consistency of the clusters. The clone selection algorithm is introduced to generate the initial code book using the splitting method. Further, it is proposed to generate the final code book using the flow distance based clone selection clustering method for optimization. The experimental results show that MDCSA achieves relatively high PSNR values for different code book sizes. The design method of clone selection based on manifold distance has better performance and effectiveness.