Quantification Vectorielle: Codes de Permutation et Quelques Techniques de Recherche Rapide (Vectorial Quantization: Permutation Coding and Some Fast Search Techniques)

摘要

Techniques for reducing the complexity of memoryless vector quantizers are develped. The simplification of usual vector quantizers such as those designed by the LBG algorithm are discussed. The use of permutation codes with simple encoding procedures and relatively small memory requirements is studied. The discussed methods were tried on Gaussian sources, speech waveforms, and LPC cepstral coefficients extracted from speech. The results are compared with full search vector quantizers. With a fairly small number of codes, such a coding system shows important performance improvement over single code for all three signals.