Blind Image Deblurring via Weighted Dark Channel Prior

Blind image deblurring is a challenging problem, which aims to estimate the blur kernel and recover the clear image from the given blurry image. A large number of image priors have been proposed to tackle this problem. Inspired by the fact that the blurring operation increases the ratio of dark channel to local maximum gradient, a weighted dark channel (WDC) prior is presented in this paper for blind image deblurring. It is shown that the WDC is more discriminative than the dark channel. The model is constructed by applying L-1 norm to the WDC term and incorporating it into the traditional deblurring framework. The alternating optimization strategy is adopted together with the half-quadratic splitting method and the fast iterative shrinkage-thresholding algorithm (FISTA) to deal with the presented model, and the maximum-minimum filter is used to improve computational efficiency. Extensive experiments are conducted on the frequently used synthetic datasets and real images, and peak signal to noise ratios (PSNR), error ratio, structural similarity (SSIM) and so on are adopted to appraise our method and some other latest methods. Qualitative and quantitative results show that our method outperforms the state-of-the-art methods.