Regularized iterative Weiner filter method for blind image deconvolution

Blind image deconvolution is a highly ill-posed problem. As a generalization of the well known Weiner filter, the existing iterative Weiner filter (IWF) method for blind image deconvolution is unstable and suffers from serious ringing artifacts. To overcome these drawbacks, in this paper, we propose two novel regularized iterative Weiner filter methods. We assume that both the latent image and the convolution kernel can be estimated by applying two different filters on the observed image. To estimate the filters, we propose to minimize energy functionals combined by the mean square errors with some regularization terms. Both H¹ and total variation (TV) regularization are considered. By applying alternating minimization method and operator splitting technique, we derive iterative algorithms for each regularization method. The proposed methods are effective for blind deconvolution of Gaussian blurred images which is widely observed in real applications such as microscopic images. Numerical experimental results on both synthetic images and real microscopic images are presented. The comparisons show that the proposed regularized algorithms perform better than the closely related state-of-the-art methods in terms of peak signal-to-noise ratio (PSNR) and visual quality.