Multi-step fixed-point proximity algorithms for solving a class of optimization problems arising from image processing (Q2348695)

The authors consider the optimization problem that appears frequently in image processing, whose objective function is the sum of two convex functions with one composed by an affine transformation. They propose a class of fixed-point multi-step proximity algorithms for solving the problem based on a fixed-point characterization of its solutions. Many existing algorithms can be identified as special cases of the proposed algorithms. The notion of weakly firmly non-expansive operators is introduced as mathematical tool for analyzing convergence of the proposed algorithms. A unified approach for convergence analysis of the new algorithms and as well as existing algorithms is provided. Convergence analysis of the algorithms can be carried out by verifying conditions on the matrices used to construct the algorithms. In particular, it is shown that the well-known alternating split Bregman iteration method is a special case of the proposed algorithm and the method is modified to improve its convergence property. A class of two-step algorithms for the total variation based image restoration models is presented.