Abstract
Image deblurring is an ill-posed inverse problem. Most algorithms make some assumptions of latent images or blur kernels. Since these assumptions may not suit for all kinds of images, they get some artifacts like ringing or over-sharpen edges with the images which do not fit to their assumptions. In this work, we propose a new post-processing algorithm for image deblurring. The algorithm modifies irregular values on the gradient domain of the latent image, and then reconstructs a refined image with the gradients. It can reduce ringing artifacts and over-sharpened edges effectively. The proposed algorithm does not make any assumption of latent images. That is, it can be applied to most types of images. Our experiments show that the method could be applied to several conventional image deblurring methods and get good results.