Lossless Image Coding Exploiting Local and Non-local Information via Probability Model Optimization

We previously proposed a lossless image coding method based on examples search and probability model optimization. In this paper, we improve coding efficiency of the method by introducing an adaptive prediction technique. Specifically, multiple affine predictors are trained pel-by-pel by using causal neighbor pels, and the predicted values obtained by those predictors are used for generating the probability model. Therefore, both non-local information by the examples search and local information by the adaptive prediction are used together in the probability modeling. Furthermore, an optimization method for the number of examples is also proposed in this paper. Experimental results show that the proposed method achieves better coding rates than the state-of-the-art lossless coding schemes.