Regularization methods for ill-conditioned linear systems

Description

The solution of linear systems coming from the discretization of Fredholm integral equations of the first kind is usually an ill-conditioned problem. Ordinary methods give unacceptable solutions, so that regularization methods are necessary. Since the systems arising from image reconstruction methods are usually large and sparse, our research is mainly focused on iterative regularization methods, and in particular conjugate gradients type regularization methods ,  that are suitable for such systems.

Main results

The study of regularization methods is mainly centred on the choice of the regularization parameter. In conjugate gradient type methods, the parameter is identified with the number of iterations of the method. Our reserach is studying some heuristic criteria based on the behaviour of the residuals and of the approximate solutions. These criteria do not need any information about the noise and hence they can be applied on any linear system, even on least squares problems. We tested these criteria on systems coming from tomographic imaging reconstruction and from dynamic MR imaging reconstruction.

Publications

·        E.Loli Piccolomini  F. Zama  G. Zanghirati, Regularization Methods in dynamic MRI ,   Applied Mathematics and Computation, September 2002.

·        M. Bertaja, S. Morigi, E. Loli Piccolomini, F. Sgallari, F. Zama, Regularization of Large Discrete ill posed problems in image processing, Recent trends in Numerical analysis (ed. Trigiante ), Advances in the Theory of Computational Mathematics, vol. 3,  Nova Science, Books and Journals (2000) (ISBN 1-56072-885-X)

·        E. Loli Piccolomini,  F. Zama, Regularization Methods for the solution of Inverse Problems: Theory and Computational Aspects, Rendiconti del circolo Matematico di Palermo, ed.  M. Maugeri, E. Galligani, serie II numero 58, 1999.

·        A. Baronio, E. Loli Piccolomini,  F. Zama, A Method for solving the Indirect Approximation Problem, Applied Mathematics and Computation, vol. 77, pag.97-107 (1996)
E. Loli Piccolomini, F. Zama, Regularization algorithms for image reconstruction from projections, Atti dell'Accademia delle Scienze di Bologna,1996.

·        E. Loli Piccolomini,  F. Zama, Regularization algorithms for image reconstruction from projections, Series on Advances in Mathematics for Applied Sciences, (edited by P. Ciarlini, M.G.Cox, F. Pavese, D. Richter), World Scientific,vol. 45, 1997.

Keywords

regularization methods, ill conditioned systems, conjugate gradients, regularization parameter