Vector quantization of image pyramids with the ECPNN algorithm
Diego P. de Garrido, William A. Pearlman, Weiler A. Finamore · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1991
A recent algorithm for single rate vector quantization [1] is used for coding image pyramids. The algorithm, called entropy-constrained pairwise-nearest-neighbor (ECPNN), designs codebooks by merging the pair of Voronoi regions which gives the most decrease in entropy for a given increase in distortion. In terms of performance in the mean-squared-error sense the algorithm produces codebooks with the same performance as the ECVQ design algorithm [1,2]. The main advantage over ECVQ is that ECPNN algorithm enables much faster codebook design. A single pass through the ECPNN design algorithm, which progresses from larger to successively smaller rates, allows the storage of any desired number of optimal intermediate-rate codebooks. In the context of pyramid coding, this feature is especially desirable, since the ECPNN design algorithm must be run for each sub-band and storage of codebooks of different rates are required for each subband. Good results at 0.5 bpp, judged both visually and using peak-to-peak SNR criterion, have been obtained by coding image pyramids using ECPNN codebooks.