Degraded Image Restoration Based On Sparsity And Dark Channel Prior

During the process of acquisition and transmission,image qualities are easy to be affected by external conditions such as illumination variation or camera shake.One of the main degradation phenomenon is blurring.Image restoration is the process that restoring the original sharp image as much as possible by developing and solving a degradation model.The existing methods usually build the energy function based on natural image priors which are difficult to simulate accurately by mathematical expression,therefore,noises and ring affects usually appear in the restored images.This paper establishes the general energy function based on Bayesian framework to solve these problems,and also builds two objective functions based on sparse and dark channel priors which make the restored result favors the sharp image.The main work and contributions of this paper are as follows:1.For the inaccurate regularization based on sparse prior,this paper builds a new and more accurate function to approximate 0L norm and works out its closed-form solution through the half quadratic splitting method.2.Due to the low constraint of sparse prior on the sharp image,this paper proposes a regularization based on both sparse and dark channel priors to constrain the restored result favors the sharp image.3.Aiming at the noise in the estimated kernel,this paper puts forward a blocking area constraint method to restrain the noise in the kernel.Finally,the proposed algorithm is tested on both Levin and Koehler public datasets and several natural blur images to illustrate the accuracy and convergence of kernel estimation through Kernel Similarity measurement.Compared with the existing algorithms based on PSNR and SSIM assessment criteria,the results of this algorithm are closer to the ground-truth visually and quantifiably.The proposed algorithm also reduces the running time significantly.