Abstract
A near-lossless image compression scheme is presented. It is essentially a DPCM system with a mechanism incorporated to minimize the entropy of the quantized prediction error sequence. With a "near-lossless" criterion of no more than a d gray level error for each pixel, where d is a small non-negative integer, trellises describing all allowable quantized prediction error sequences are constructed. A set of "contexts" is defined for the conditioning prediction error model and an algorithm that produces minimum entropy conditioned on the contexts is presented. Finally, experimental results are given.