this paper we consider the problem of separating noisy instantaneous linearmixtures of document images in the Bayesian framework. The source image ismodeled hierarchically by a latent labeling process representing the commonclassifications of document objects among different color channels and theintensity process of pixels given the class labels. A Potts Markov random fieldis used to model regional regularity of the classification labels inside objectregions. Local dependency between neighboring pixels can also be accounted bysmoothness constraint on their intensities. Within the Bayesian approach, allunknowns including the source, the classification, the mixing coefficients andthe distribution parameters of these variables are estimated from theirposterior laws. The corresponding Bayesian computations are done by MCMCsampling algorithm. Results from experiments on synthetic and real imagemixtures are presented to illustrate the performance of the proposed method.
展开▼
