Bayesian blind separation of mixed text patterns

In this paper we consider the problem of unsupervised separation of mixed text patterns based on blind source separation models. We propose a hierarchical Markov random field model for the source patterns, which enforces piece-wise regularity on both labels and intensities of image pixels. We also presented a hierarchical Bayesian BSS framework, in which the unknown sources and labels is estimated through a generic iterative algorithm framework on the basis of corresponding posterior laws. Experiment results on synthetic and real sample images are presented to show the feasibility of the proposed model.