Design of a minimum-rate predictor and its application to lossless image coding

In this paper, we propose a novel method for designing linear predictors suitable for lossless image coding. In general, the Minimum Mean Square Error (MMSE) is an important concept in image coding. However, predictors which are optimized on the basis of the MSE criterion are not necessarily optimum from a viewpoint of coding efficiency. Thereupon our method optimizes a predictor so that a cost function which represents an amount of information on prediction errors can have a minimum. Moreover, we develop an adaptive lossless coding scheme for still images to demonstrate effectiveness of the proposed method. Simulation results indicate that the proposed coding scheme is superior in terms of coding efficiency to the conventional scheme which utilizes the MMSE predictors, and that a coding rate of the proposed scheme is 0.14–0.40 bits/pel lower than that of the JPEG-LS standard coding scheme.