Fast search for entropy-constrained VQ

The design of fast nearest-neighbor algorithms is a crucial issue in vector quantizers using unstructured codebooks. This paper presents a fast linear search algorithm for entropy-constrained vector quantizers, which use a modified distortion measure taking into account the entropy of the output. It consists in an adaptation of previous methods designed for standard vector quantizers. New elimination rules are described, based on the intensity mean and variance of the encoded blocks, and allowing a dramatic reduction of the search space at the expense of a few more comparisons. Experimental results on a set of greyscale images are presented.