基于核估计迭代结构保持的图像盲复原方法

Blind image de-blurring based on structure preserving in iterative kernel estimation

振铃抑制问题是鲁棒的模糊核估计中的关键问题。针对该问题,本文提出根据核估计迭代中潜在图像内振铃和真实图像结构的变化特性对二者进行区分。根据观察,潜在图像内的真实图像结构能够在迭代中稳定保持,振铃却随着核估计而不断变化。在本文所提方法中,我们利用迭代过程中潜在图像的序列来提取对于核估计有利的结构。实验结果表明,本文所提出方法相比现有的图像盲复原方法,在仿真图像数据和真实图像数据上表现更加优异,算法更加鲁棒。

Ringing suppression is a key issue for robust blur kernel estimation. On this problem,this paper propose to distinguish ringing effect and actual structures based on their characteristics during iteration. According to our observation,real structures in a latent image can be maintained stably while the ringing keeps changing when the kernel being optimized in iterations. We exploit gradients during the iteration sequence as the principle to extract favorable image structures for accurate kernel estimation. Extensive experimental studies demonstrate that the proposed method is more robust and performs favorably against the state-of-the-art image deblurring methods on both synthesized and natural images.