Genetic clustering algorithm for mean-residual vector quantization

Vector quantization is a useful tool for data compression. It can be applied to compress the data vectors in the database. The quality of the recovered data vector depends on a good codebook. In this paper, the genetic algorithm combined with the generalized Lloyd algorithm (GLA) is applied to the codebook design of mean-residual vector quantization (M/R VQ). The mean codebook and residual codebook are trained using GLA algorithm separately, then the genetic algorithm is used to evaluate and evolve the combined mean codebook and residual codebook. Experimental results demonstrate the proposed genetic clustering algorithm applied to M/R VQ may improve the peak signal to noise ratio of the recovered data vector by comparing with the GLA algorithm.