CODEBOOK DESIGN FOR VECTOR QUANTIZATION USING GENETIC ALGORITHM

Genetic algorithms have been widely used to solve optimization in many fields such as multi-objective optimization, Fuzzy Optimization, and scheduling problem. Vector quantization, a basic method, is adopted by image compression technology and has a better performance than scalar quantization. Hence, it is worth to study how to apply genetic algorithms on the optimal design of codebook generation in vector quantization, where a codebook could minimize the average distortion between a given training set and the codebook. The advantage of the traditional genetic algorithms will be used in this paper, different from LBG training method, to evolve out a better codebook which is the nearest representative one by a fitness function in vector quantization. We take the random combinations of codebooks of the training samples as the initial population. Peak Signal-to-Noise Ratio is used as the fitness value. Using Two-point crossover and mutation process will get a better codebook finally after the evolution iterations in the proposed paper. Proved by our experiment, the proposed method, Simply Genetic Codebook Algorithm, can evolve and generate a better codebook through the whole experiment. To generate and acquire the high-quality codebook, the efficiency of codebook generation will be considered improved by imbedding stochastic model in the future.