Blind image deblurring algorithms have been improving steadily in the pastyears. Most state-of-the-art algorithms, however, still cannot performperfectly in challenging cases, especially in large blur setting. In thispaper, we focus on how to estimate a good kernel estimate from a single blurredimage based on the image structure. We found that image details caused byblurring could adversely affect the kernel estimation, especially when the blurkernel is large. One effective way to eliminate these details is to apply imagedenoising model based on the Total Variation (TV). First, we developed a novelmethod for computing image structures based on TV model, such that thestructures undermining the kernel estimation will be removed. Second, tomitigate the possible adverse effect of salient edges and improve therobustness of kernel estimation, we applied a gradient selection method. Third,we proposed a novel kernel estimation method, which is capable of preservingthe continuity and sparsity of the kernel and reducing the noises. Finally, wedeveloped an adaptive weighted spatial prior, for the purpose of preservingsharp edges in latent image restoration. The effectiveness of our method isdemonstrated by experiments on various kinds of challenging examples.
展开▼
