Alternative methods for codebook design in vector quantization

Abstract

A vector quantizer maps a multidimensional vector space into a finite subset of reproduction vectors called a codebook. For codebook optimization the well known LBG algorithm or a simulated annealing technique are commonly used. Two alternative methods the fuzzy-c-mean (FCM) and a genetic algorithm (GA) are proposed. In order to illustrate the algorithm performance a DCT-VQ has been chosen. The fixed partition scheme based on the mean energy per coefficient is shown for the test image "Lena".