Vector quantization(VQ) has proven to be an efficient technique for commpression of a sampled image at under 1.0 bpp. VQs in the pixel domain can be defined as a kind of pattern-matching coder which replaces a pixel subblock by some matching pattern. In the VQ, the pattern set(codebook), a collection of possible matching patterns, is generated from training images in advance. If the statistical properties of compressed image and training images are not matched, the problem of pattern-set mismatch occurs and excess quantization error results. To solve this problem, this paper develops a novel universal pattern-matching scheme, named PMC for short,which has the power of self-organizing its own pattern-set and which basically escapes the pattern-set mismatch.
View full abstract