Codebook Design for Vector Quantization with Most Dispersed Codewords in Initialization

Many improvements on LBG algorithm are achieved at the expense of more runtime.This paper presents a novel improvement on codebook design for image vector quantization with the most dispersed codewords in initialization(MDCI).In MDCI,all the initial codewords are selected from the inputted training vectors set,and the distance between the next newly generated initial codeword and the already existed codewords must be the greatest.Experimental results demonstrate MDCI conquers the empty cell problem,alleviates the problem of local optima more effectively and gets higher-performance codebook,outperforming the conventional LBG algorithm and many LBG-based modified algorithms,like the ant colony optimization based codebook design algorithm with respect to both codebook performance and runtime.