A fast evolutionary algorithm in codebook design
Abdolali Momenai, Siamak Talebi · International Conference on Artificial Intelligence · 2005
This paper presents a fast algorithm to find an optimal subset codebook from a super codebook in a way that the RMS error between the new codebook and the training set in vector quantization becomes minimum. To have a fast algorithm, a genetic based algorithm is used that uses 2 evolutions, one in designing the whole sub-codebook and the other in finding each individual codeword of the sub-codebook. This optimal codebook can be efficiently used in the real time compression of the images with PSNR improvement of about 1.1 - 8.2db in blocks of the image.