Modified Markov Random Fields-based Variational Bayesian Imaging Approach for Cluster Structured Faint Scattered Targets

The image reconstruction of the traditional Markov Random Fields (MRF) -based Bayesian compressive sensing (BCS) method suffers from noise, especially for the faint targets. In this paper, a modified MRF-based BCS imaging approach is proposed to deal with these issues. For the de-noising purpose, a threshold is introduced in the Bayesian clustered sparse model. In addition, the variational Bayesian expectation-maximization (VBEM) is an iterative algorithm for solving the BCS optimization problem, hence it tends to converge to a local optimum, which will easily lead to the loss of weak scatterers. To surmount this difficulty, an average filter is performed on the support vector in each iteration. The simulation and experimental results demonstrate that the proposed method can achieve substantial improvements over the original MRF-based BCS method in terms of noise suppression and reconstruction of faint targets.