COMPARISON OF REGULARIZED AND SUPERIORIZED METHODS FOR TOMOGRAPHIC IMAGE RECONSTRUCTION.

We compare two approaches to image reconstruction in computed tomography (CT) which incorporate penalty functions to improve image quality in the presence of noisy data. The first approach adapts a previously proposed hybrid method for solving a regularized least squares problem, which simultaneously computes the regularization parameter and the corresponding solution. The second approach is based on the superiorization methodology, wherein the solution is perturbed between iterations of a feasibility-seeking algorithm to minimize a secondary objective. Numerical experiments indicate that while both approaches are able to significantly improve image quality, the heuristic applied to select the regularization parameter in the hybrid method does not generalize well to the CT reconstruction problem. The superiorization methodology is more effective, provided that a suitable stopping criterion can be determined. [ABSTRACT FROM AUTHOR]