Efficient implementation of image recovery algorithms

Abstract

Presents an efficient implementation scheme for the Newton algorithm for convex set constrained signal recovery [Dharanipragada and Arun, 1993). The implementation avoids matrix creation, inversion and even storage by exploiting the structure of the operators involved and by using conjugate-gradient iterations within each Newton iteration. The resulting algorithm is thus computation and memory efficient and is well-suited for large scale problems typically arising in image recovery applications. The same implementation scheme can also be effectively used for the POCS algorithm.<>