Choosing the Parameter of Regularized MP-Inverses by Modified SIC in Image Restoration Problems

Image restoration is a typical ill-posed problem. Regularization technic represented by a regularized MP-Inverse filter (RMPIF) is widely used to deal with the ill-posedness. In order to derive the best performance of the filter, the parameter, which controls the regularizability, should be appropriately chosen.

In this paper, we present a new criterion for parameter choosing based on modifying the subspace information criterion which is first proposed for model selection of supervised learning problems. Some numerical examples are also shown to verify the efficacy of the proposed criterion.