A New Hyper-Laplacian Prior-Based Deconvolution Method for Single Image Deblurring

Motion blur is an ill-posed problem which has been addressed for a long time in digital photographing. It usually happens under the circumstance without enough surrounding light. Several methods have been proposed for this problem. However, the ringing is inevitable artifacts arising in the deconvolution stage. To suppress undesirable artifacts, regularization-based methods have been proposed, and they use the natural image priors to overcome the ill-posedness of deconvolution problems. These studies suggest that priors based on natural image statistics can regularize deblurring problems to yield better results. In this paper, we propose a new single image deblurring algorithm based on the hyper-Laplacian priors. We apply hyper-Laplacian priors to obtain a precise blur kernel in the point spread function (PSF) estimation phase. Our proposed method successfully reduces ringing artifacts while preserving the edge of images. Experimental results simulated and real images show that our method performs better than other methods.