Abstract
The method of projections onto convex sets can be used to solve many problems in image restoration, e.g., restoration from phase, spectral extrapolation, and signal recovery in computer-aided tomography. However, image-restoration problems involving nonconvex constraints cannot be handled by the method of projection onto convex sets in a fashion that ensures convergence. The restoration-from-magnitude (RFM) problem is such a case. To handle the RFM as well as other nonconvex constraints, we describe an algorithm known as generalized projections and discuss its properties. When sets are nonconvex, it is possible for the algorithm to exhibit pathological behavior that is never manifest in convex projections. We introduce an error criterion called the summed-distance error (SDE) and show under what circumstances the SDE is a monotonically decreasing function of the number of iterations. Near-optimum performance of the algorithm is achieved by relaxation parameters. Comparisons with other RFM methods are furnished for synthetic imagery.
© 1984 Optical Society of America
Full Article | PDF ArticleMore Like This
Eitan Yudilevich, Aharon Levi, G. J. Habetler, and Henry Stark
J. Opt. Soc. Am. A 4(1) 236-246 (1987)
Heinz H. Bauschke, Patrick L. Combettes, and D. Russell Luke
J. Opt. Soc. Am. A 20(6) 1025-1034 (2003)
Henry Stark and Elwood T. Olsen
J. Opt. Soc. Am. A 9(11) 1914-1919 (1992)