A blur-sure-let algorithm to blind PSF estimation for deconvolution
Feng Xue, Jiaqi Liu, Chengguo Liu, Gang Meng · 2014
In this paper, we consider blind deconvolution that consists of PSF (point spread function) estimation and non-blind deconvolution. It has been proved that blur-SURE - a modified version of SURE (Stein's unbiased risk estimate) - is a valid criterion for parametric PSF estimation. The key contribution of this work is to propose a fast algorithm for the blur-SURE minimization. Incorporating a linear combination of multiple smoother matrices with different but fixed regularization parameters, the optimal regularized processing is obtained by a closed-form solution. This linear parametrization of the processing greatly accelerates the minimization. The extensive experiments show the significant improvement of computational time. The low computational complexity enables the blur-SURE criterion to be readily applied to more complicated parametric forms of PSF.