An Algorithm-Independent Analysis of the Quality of Images Produced Using Multi-Frame Blind Deconvolution Algorithms--Conference Proceedings (Postprint)
Matson, Charles, Alim Haji · Defense Technical Information Center (DTIC) · 2007
Multi-frame blind deconvolution (MFBD) algorithms can be used to generate a deblurred image of an object from a sequence of short-exposure and atmospherically-blurred images of the object by jointly estimating the common object and all the blurring functions. In this paper we present fundamental limits (Cramer-Rao lower bounds) to the quality of restored images generated with MFBD algorithms. We show that image restorations are less noisy when using a Zemike-based point spread function (PSF) parameterization than when using a pixel-based PSF parameterization. We also show that the most noise reduction tends to occur near the edges of the true object support, even when a larger support region is used as a constraint. Finally, we show that Zernike-based PSF parameterization produces higher resolution in restored images than does pixel-based PSF parameterization.