Blind image deconvolution using the Gaussian scale mixture fields of experts prior

In this paper, a blind image deconvolution method which is derived from Bayesian probabilistic framework is proposed. A robust prior named Gaussian Scale Mixture Fields of Experts (GSM FoE) and a prior that is constructed with the lp-norm (p ≈ 1.5) are adopted to regularize the latent image and the point spread function (PSF) respectively. We use a two phase optimization approach to solve the resulted maximum a-posteriori (MAP) estimation problem, and a simple gradient selecting method is incorporated into the alternating minimization to improve the accuracy of the estimated PSF. Experiments on both synthetic and real world blurred images show that our method can achieve results with high quality.