An algorithmic study on context modelling for lossless image compression

Statistical context modelling is a powerful and versatile technique for lossless image coding. A key issue in context modelling is how to increase the order of model without drastically increasing the model cost. We take an algorithmic approach to address the issue and propose a few heuristical optimization techniques that fine tune models of relatively few parameters in the sense of entropy minimization. Our experimental results indicate that these techniques are quite effective and achieve the lowest lossless bit rates so far over a variety of test images.