This work investigates a simple yet effective image deblurring method that combines salient image edges with structures. Typically, the vast majority of edge-based work focuses on advancing the salient edges information for blind image deblurring while ignoring the refined scale of the image structures. With this in mind, we show how to effectively exploit the image structure to blind image deblurring in this paper, i.e., we adopt a mutually guided image filter to guide the restoration of the image structure while salient edges provide edge information for blur kernel estimation. Thus, a more expressive deblurring model that employs ${\mathit{L}}_2$-regularization for salient edges and image structure, together with the ${\mathit{L}}_0$-regularization for image gradients is proposed. We find that the quality of the recovered kernel is thereby improved, and the deblurring results are more satisfactory. Experimental results show that our method outperforms state-of-the-art methods in terms of benchmark-based datasets and real scenarios, as well as computational efficiency.

这项工作研究了一种简单而有效的图像去模糊方法，该方法将显着图像边缘与结构相结合。通常，绝大多数基于边缘的工作都集中在提升显着边缘信息以进行盲图像去模糊，而忽略了图像结构的精细尺度。考虑到这一点，我们在本文中展示了如何有效地利用图像结构进行盲图像去模糊，即我们采用相互引导的图像滤波器来指导图像结构的恢复，而显着边缘为模糊核估计提供边缘信息。因此，一个更具表现力的去模糊模型采用${\mathit{大号}}_2$- 显着边缘和图像结构的正则化，以及${\mathit{大号}}_0$- 提出了图像梯度的正则化。我们发现，由此提高了恢复内核的质量，并且去模糊效果更令人满意。实验结果表明，我们的方法在基于基准的数据集和真实场景以及计算效率方面优于最先进的方法。