The University of British Columbia

This paper presents an analysis and compression technique that can be used for both lossy and lossless compression of black and white documents simultaneously. It is assumed that an application-dependent analysis technique is employed to produce weighting coefficients that allow ordering of the bits according to their significance. Like the algorithm described in the JBIG standard, the proposed algorithm consists of high order statistical modeling and adaptive arithmetic coding. However, our modeling techniques are more sophisticated in the sense that they are adaptive both locally and globally. The conditioning region of support used for the generation of the states is determined based on the global statistics of the input image, and the states and associated probabilities are adapted to the local statistics. Moreover, our algorithm is naturally suitable for progressive transmission since the output bit stream can be truncated anywhere, leading to the best possible approximation given a bandwidth constraint. Experimental results reveal that the proposed algorithm not only achieves high near-lossless compression performance but also outperforms JBIG when used for lossless compression