Quantização vetorial aplicada à compressão de sinais de voz e imagem.

Abstract

Vector quantization has been widely used as an excellent technique for signal compression, specially voice and image. I n this work two techniques for vector quantizers design are presented. The first technique concerns the introduction of some modifications in Kohonen's neural network training algorithm. The second concerns the derivation of an algorithm based on Principal Component Analysis. The performance of each technique is compared to the performance presented by the traditional LBG algorithm and the superiority of the proposed algorithms have been evidenced. For evaluating the great potential and efficiency of vector quantization, results of simulations of voice signals, images and signals with known distributions are presented.