A regularized dual-based iterative method for a class of image reconstruction problems

An iterative method for a class of image reconstruction problems which lead to large scale optimization problems is presented. The method uses a regularization of the objective functional and is based on its dual formulation which is a semi-separable convex minimization problem with linear constraints, where the function to be minimized is the sum of a Burg's entropy and a quadratic function. From the special structure of this new formulation in combination with a Bregman type method, a computationally attractive algorithm emerges and its convergence properties are proved.