An estimator for image desmearing using a Bernoulli-Gaussian model

The problem of deconvolving severely smeared images is considered in this paper. It is shown that classical techniques for deconvolution are, for some data sets, inadequate due to the severity of the smearing. The reasons for this are investigated and a measure is obtained for the signal to noise ratio improvement of estimators operating on isotropic smearing on a circular domain. The issue of what can be estimated from such data sets is then investigated and an adaption of the Kormylo-Mendel single most likely replacement algorithm is proposed as a method of estimating sparse sources from such data sets.