Evaluation of the biased run-length coding method on binary images generated by a modified Ising model

In this work, we evaluate the performance of several lossless compression methods on binary images simulated by a statistical model. This work was motivated by our previous study on compression of regions-of-interest maps in hyperspectral images, where a biased run-length coding method was found to provide good compression on just a few binary images with most pixels being in the background. In order to discover the trend on the compression performance, we introduced a modified Ising model of Markov random fields by Metropolis, which can generate a large number of binary images iteratively, so that we can study how the compression ratios of the above method change gradually when the contents of the images get varied slightly with iterations. Simulation results showed that the biased run-length coding method significantly outperformed the arithmetic code and JBIG2 method. The study also gave rise to a dual biased run-length method, which can provide further compression gains on images with more foreground objects.