Robust deblurring of random blur

Abstract

A robust method for deblurring random blur is introduced. The image is assumed to be distorted by a linear system whose impulse response function is itself random and by additive noise whose spectral density is known only to be in the neighborhood of some specified spectral density. The estimate is motivated from the generalized least-squares and robust statistical methods. Our robust deblurring is considered in the frequency domain and is of the form of weighted least squares, with the most prominent frequencies of the random impulse response being downweighted in a way similar to Huber's robust estimator.