基于加权平均曲率正则化的盲图像去模糊方法

文章将加权平均曲率正则化项作为隐式边缘选择项，结合相互引导图像滤波实现显式边缘选择的策略，提出一种基于加权平均曲率正则化的盲图像去模糊方法。该方法采用加权平均曲率正则化隐式去除不利结构和中间潜在图像中的细节并保留显著边缘，有利于模糊核估计；在求解去模糊模型时，引入半二次分裂方法和加权平均曲率滤波来优化算法，有效地求解模型。结果表明，该方法在基准数据集和真实的模糊图像上均优于其他去模糊方法。

In this paper, a blind image deblurring method based on weighted mean curvature regularization is proposed, which combines explicit and implicit salient edge selection strategies, that is, taking weighted mean curvature regularization term as implicit edge selection term, and using mutually guided image filtering to achieve explicit edge selection. The weighted mean curvature regularization method is used to remove the unfavorable structure and the details in the intermediate potential image, and preserve the salient edges, which is conducive to blur kernel estimation. In addition, the half-quadratic splitting method and weighted mean curvature filter are used for algorithm optimization to approximate the model effectively. Experiments show that the proposed method is superior to other blind deblurring methods on both benchmark datasets and real-world blurry images.