Fast implementation of two-level compression method using QM-coder

Abstract

We deal with bi-level image compression. Modern methods consider the bi-level image as a high order Markovian source, and by exploiting this characteristic, can attain better performance. At a first glance, the increasing of the order of the Markovian model in the modelling process should yield a higher compression ratio, but in fact, it is not true. A higher order model needs a longer time to learn (adaptively) the statistical characteristic of the source. If the source sequence, or the bi-level image in this case, is not long enough, then we do not have a stable model. One simple way to solve this problem is the two-level method. We consider the implementation aspects of this method. Instead of using the general arithmetic coder, an obvious alternative is using the QM-coder, thus reducing the memory used and increasing the execution speed. We discuss some possible heuristics to increase the performance. Experimental results obtained with the ITU-T test images are given.