Lossless Coding Using Predictors and Arithmetic Code Optimized for Each Image

This paper proposes an efficient lossless coding scheme for still images. The scheme utilizes a block-adaptive prediction technique to effectively remove redundancy in a given image. The resulting prediction errors are encoded using a kind of context-adaptive arithmetic coding method. In order to improve coding efficiency, a generalized Gaussian function is used as a probability distribution model of the prediction errors in each context. Moreover, not only the predictors but also parameters of the probability distribution models are iteratively optimized for each image so that a coding rate of the prediction errors can have a minimum. Experimental results show that an average coding rate of the proposed coding scheme is close to 90% of that of JPEG-LS and is lower than that of TMW.