Talk by Marco Prato, Università di Modena e Reggio Emilia

From ISRA and EM to variable metric inexact linesearch algorithms: application to imaging problems

Abstract

Many problems in image processing can be reformulated as the minimization of a functional given by a smooth äóñ possibly nonconvex äóñ data fidelity term plus convex äóñ possibly nonsmooth äóñ regularization terms, including indicator functions of convex sets when constraints are available on the desired images. Starting from the iterative space reconstruction algorithm and the expectation maximization method, we show the successive generalizations to the split gradient, scaled gradient projection and variable metric inexact linesearch algorithms, the last one being a recent approach able to address very general formulations of imaging problems in a mathematically sound and practically efficient way.