An extended maximum likelihood approach for the robust blind separation of autocorrelated images from noisy mixtures

In this paper we consider the problem of separating autocorrelated source images from linear mixtures with unknown coefficients, in presence of even significant noise. Assuming the statistical independence of the sources, we formulate the problem in a Bayesian estimation framework, and describe local correlation within the individual source images through the use of suitable Gibbs priors, accounting also for well-behaved edges in the images. Based on an extension of the Maximum Likelihood approach to ICA, we derive an algorithm for recovering the mixing matrix that makes the estimated sources fit the known properties of the original sources. The preliminary experimental results on synthetic mixtures showed that a significant robustness against noise, both stationary and non-stationary, can be achieved even by using generic autocorrelation models.