Codebook optimization by Cauchy annealing

The codebook design is a central issue of vector quantisation (VQ) for speech or image signal compression. Simulated annealing (SA) has been applied to codebook design with encoder perturbations. This paper presents a vector quantizer design method which associates the LBG algorithm with Cauchy annealing (CA) by decoder perturbations. This stochastic algorithm is able to find a globally optimal solution much faster than classical simulated annealing methods and remove the initial condition as the codebook design parameter. Moreover, deterministic Cauchy annealing algorithms have been proposed for computing complexity reduction and they are much more efficient than stochastic Cauchy annealing.<>