Blind image deblurring based on dictionary replacing

Abstract

Traditional image deblurring is based on deconvolution, an ill-posed problem, which is sensitive to the accuracy of the blur kernel. In this paper, we propose a blind image deblurring method based on dictionary replacing. First, we estimate the blur kernel from the blur image , and then based on the sparse representation of the image patch under over-complete dictionary, we deblur the image via replacing blur dictionary with clear dictionary. Our method avoids the deconvolution problem and can bring more high-frequency information in the deblurred image via dictionary replacing. Experimental results compared with state-of-the-art blind deblurring methods demonstrate the effectiveness of the proposed method. © 2012 Springer-Verlag.