Image Reconstruction with Component Regularization Based on Bregman Iteration

Regularization technique is a common method for solving inverse problems in image processing. For a complex natural image which contains various structure components, it may be more effective to adopt different regularization terms for different components. An optimization model with component regularization, which was used to solve the problem of compressed sensing image reconstruction, has been established by us. In this paper, general linear inverse problems in imaging are considered, and an alternating iterative algorithm based on Bregman iteration is proposed to solve the optimization problem with component regularization. This iterative algorithm is applied to reconstruct natural images which contain piecewise smooth and texture components, in the compressed sensing framework. Experimental results show the effectiveness of the proposed algorithm.