An EM approach for Poisson-Gaussian noise modeling

This paper deals with noise parameter estimation. We assume observations corrupted by noise modelled as a sum of two random processes: one Poisson and the other a (nonzero mean) Gaussian. Such problems arise in various applications, e.g. in astronomy and confocal microscopy imaging. To estimate noise parameters, we propose an iterative algorithm based on an Expectation-Maximization approach. This allows us to jointly estimate the scale parameter of the Poisson component and the mean and variance of the Gaussian one. Moreover, an adequate initialization based on cumulants is provided. Numerical difficulties arising from the procedure are also addressed. To validate the proposed method in terms of accuracy and robustness, tests are performed on synthetic data. The good performance of the method is also demonstrated in a denoising experiment on real data.