Multichannel image restoration using combined channel information and robust M-estimator approach

We propose a novel multichannel image restoration scheme based on Bayesian maximum a posteriori estimation. Restoration schemes for mono-channel images fail in the case of multichannel images since they do not take into account the cross channel relations present. We propose a robust M-estimator based Gibbs prior which uses not only spatial but also interchannel term for multichannel image data modelled as a Markov random field. Regularized total variation potential function is used as a robust M-estimator in each channel and a weighted inter-channel coupling term with Gaussian prior is used to capture the cross-correlations of the given image. We prove the well-posedness of the corresponding minimization in the bounded variation space. This variational Bayesian scheme reduces the channel mixup artifacts and preserves color edges when compared with other schemes on real and noisy multidimensional images. © 2010 by IJTS.