Abstract

The multiframe resolution enhancement problem (and its close cousin the single-frame “superresolution” problem) have recently received a great deal of attention. Earlier controversy as to the possibility of producing a higher resolution image from a give set of low-resolution images having been more or less settled, the key problems being studied vigorously now are related to specific approaches and implementations. What we report in this paper is progress toward the efficient and stable implementation of an algorithm for recovering high resolution images from a set of lower resolution, sub-pixel shifted frames of the same scene. The “micro-scanned” images of the scene typically form an underdetermined system of equations, that is in any case severly ill-conditioned. We propose methods for efficiently stabilizing, regularizing, and computing the solution; present numerical methods for quickly computing the regularization parameter, and for recovering the (possibly unknown) point-spread function of the imaging system

© 2001 Optical Society of America