A FAST PNN DESIGN ALGORITHM FOR ENTROPY-CONSTRAINED RESIDUAL VECTOR QUANTIZATION

A clustering algorithm based on the pairwise nearest-neighbor algorith m developed by Equitz, is introduced for the design of entropy-constra ined residual vector quantizers. The algorithm designs residual vector quantization codebooks by merging the pair of stage clusters that min imizes the increase in overall distortion subject to a given decrease in entropy. Image coding experiments show that the clustering design a lgorithm typically results in more than a 200:1 reduction in design ti me relative to the standard iterative entropy-constrained residual vec tor quantization algorithm while introducing only small additional dis tortion. Multipath searching over the sequence of merges is also inves tigated and shown experimentally to slightly improve rate-distortion p erformance, The proposed algorithm can be used alone or can be followe d by the iterative algorithm to improve the reproduction quality at th e same bit rate.