Fast Blind Deblurring via Normalized Sparsity Prior

摘要

Image deblurring is a longstanding problem in digital image processing and computer vision community. In this paper, we propose a new single image blind deconvolution method. For kernel estimation, salient edges are extracted from blurred image to increase the robustness of kernel estimation. To preserve the natural properties of latent images, we employ a powerful natural image prior to guide the latent image restoration. An efficient and fast optimization method is proposed to solve our proposed model. We have extensively tested our algorithm, and found that is able to provide correct kernel estimates and sharp images.