Blind deconvolution using least squares minimisation

Blind deconvolution of blurred images has been demonstrated by a number of authors. The quantitative performance of these algorithms is less well known, in particular the sensitivity of the reconstruction to the inherent ambiguities in blind deconvolution and the effect of noise. We consider in detail two algorithms based on a least squares optimisation approach. The performance of these two algorithms is also discussed with regard to superresolving an image corrupted by an unknown blurring function, using an example recently published in the literature [J. Opt. Soc. Am. A 11 (1994) 2401].