Efficient codebook design in vector quantization by parallel simulated annealing and evolutionary selection

Abstract

Vector quantization is a technique that has been investigated and used in speech and image coding for data compression. A hybrid system of parallel simulated annealing and evolutionary selection (PSAES) is introduced to codebook design in vector quantization. It is examined for the vector space of low-order DCT coefficients and compared with local and near global optimization techniques. In addition an improved version of a fuzzy codebook design called fuzzy-to-hard-c-mean algorithm is proposed.