Generalized Cross-Validation as a Stopping Rule for the Richardson-Lucy Algorithm

Kevin M. Perry and Stanley J. Reeves

This paper presents a criterion for stopping non-linear iterative algorithms, specifically the Richardson-Lucy algorithm that is widely used to restore images from the Hubble Space Telescope. The criterion is based on generalized cross-validation and is also computed iteratively. We will present examples displaying the power of the stopping rule, and will discuss the abilities and shortcomings of this method. We also present a caveat about the method.

Pages 97-103

Full paper available in HTML and compressed PostScript