Blind Image Deblurring via L2-Regularization

Abstract

This research presents a novel and efficient approach for deblurring the blurred image. The proposed method is based on the iterative concept and blind deconvolution which is an ill-posed problem. The clear image and blur kernel that generated the blurred image, have many pairs. It requires regularization to get an accurate answer. Our work proposes new L2 � regularization prior based on the sum of gradient with a suitable bound for solution of a deblurring problem. The method will adjust the values of pixels in the blurred image according to the suitable bound in each iteration. Furthermore, our work used an optimization algorithm based on a quadratic splitting technique that guarantees that each sub-problem approach to solution and fast convergence. The experiments on data set of images present that our approach has more effectiveness than Fergus et al. (2006) method. Moreover, our method achieves image performance results similar to Cho and Lee (2009), Xu et al. (2010), Shan et al. (2008), and Krishnan et al. (2009) method on blurry image with low light diffusion.

?????????????????????????????????????????????????????????????????? ???????????????????????????????????????????????????????????? ??????????????????????????????????? ???????????????????????????????????????????????? ????????????????????????????????????????????????????????????? ?????????????????????????????????????????????? ??????????????????????????????????????????????????? ?????????????????????????????????????????????????????????????????????? ????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????? ????????????????????????????????????? ???????????????????????????????????????????????????????????????? Fergus et al. (2006) ????????? ?????????????????????????????????????????????????????????????????? Cho and Lee (2009), Xu et al. (2010), Shan et al. (2008) ??? Krishnan et al. (2009) ?????????????????????????????????