The maximum a posteriori estimation model for signal recovery with mixed Gaussian and impulse noise (Q6066883)

The main contribution of this paper is to propose a variational model for recovering signals corrupted by mixed Gaussian and random-valued (RV) noise. First, the main difficulty in removing this mixed noise lies in modeling the nonlinear form of the noise, the authors adopt a spike-and-slab distribution to model the mixed noise so that a principled additive alternative to the nonlinear model is provided. Then, a joint maximum a posteriori (MAP) estimation for the unobservable signal and location indicators of RV noise is derived. Moreover, the variational model that only computes the MAP estimation of the signal is also presented, and the resulting data fidelity term to be minimized is the Moreau-Yosida regularization of the \(l_0\) norm. Finally, numerical experiments demonstrate that this data fidelity term exhibits superior fitting to the likelihood function for mixed Gaussian and impulse noise, in contrast with several widely used data fidelity terms. The research work is innovative; its derivation is strict and its performance is good. It would have been better to discuss the parameters of the model and experiments.