Prediction based on backward adaptive recognition of local texture orientation and Poisson statistical model for lossless/near-lossless image compression

This paper is devoted to prediction-based lossless/near-lossless image compression algorithm. Within this framework, there are three modules, including prediction model, statistical model and entropy coding. This paper focuses on the former two, and puts forward two new methods: prediction model based on backward adaptive recognition of local texture orientation (BAROLTO), and Poisson statistical model. As far as we know, BAROLTO is the best predictor in efficiency. The Poisson model is designed to avoid the context dilution to some extent and make use of a large neighborhood; therefore, we can capture more local correlation. Experiments show that our compression system (BP) based on BAROLTO prediction and Poisson model outperforms the products of IBM and HP significantly.