Abstract
This paper proposes a novel method for designing linear predictors suitable for lossless coding of still images. In general, the minimum-MSE criterion has been regarded as proper for predictive coding as well, but it is not necessarily optimum from a viewpoint of coding efficiency in lossless coding. Thereupon a predictor is optimized in our method so that a coding rate of prediction errors, instead of MSE, can have a minimum. Besides in order to improve coding efficiency much more, an image is partitioned into square blocks and these blocks are classified into some classes. Then the above-mentioned optimization of a predictor is carried out individually in each class. Simulation results indicate that the proposed scheme is superior to the conventional scheme which utilizes MSE-minimized predictors in terms of coding rates.
