Context-dependent multilevel pattern matching for lossless image compression

| The performance of the multilevel pattern matching (MPM) code for lossless image compression is rst analyzed. It is shown that the worst-case redundancy of the MPM code against all nite 2D context arithmetic codes is O(1= p log n), where n is the number of pixels in the image to be compressed. This result is in contrast to the redundancy of O(1= log n) in the case of 1D data and is caused by the so-called 2D boundary e ect. To alleviate the 2D boundary e ect, we then extend the MPM code to the case of context modeling, yielding a context-dependentMPM code. Our experimental results show that the contextdependent MPM code, with a simple context model, signi cantly outperforms the original MPM code on a wide range of bi-level images. Comparing with JBIG, the context-dependent MPM code outperforms JBIG in progressive coding mode, and is comparable with JBIG in sequential coding mode.