基于模糊图像的退化过程、卷积模糊模型和模糊图像生成的机理, 提出一种基于L0范数的正则化模糊核估计方法, 解决了遥感图像重建问题中0范数难求解的难题。该方法以模糊核稀疏性为先验知识, 采用对应梯度的L0范数为正则项, 有效避免了细小边缘对模糊核估计的影响, 使得模糊核的估计更加准确。进一步采用超拉普拉斯分布来近似图像梯度的重尾分布, 利用L0.5范数正则化对模糊图像做反卷积, 从而恢复出原始图像。与传统方法相比, 本文方法可以准确地估计出图像的模糊核, 很好地抑制恢复图像的振铃现象, 有效地提升遥感图像的质量。模糊图像以及各方法重构图像在同一刀刃下的调制传递函数(MTF)曲线显示, 本文方法的MTF曲线得到了较好的提升。

On the basis of the degradation process of a blurred image, a convolution fuzzy model and the fuzzy image generation mechanism, a zero-norm regularization kernel estimation method is proposed to overcome the problem that 0 norm is difficult to solve in the remote sensing image reconstruction. By taking a fuzzy nuclear sparse for prior knowledge and corresponding gradient norms for regular items, the method avoids the impact of small edges of the image on blurred kernel and accurately estimates the blur kernel by the blurring image. Furthermore, the super Laplace distribution is used to approximate the heavy-tailed distribution of image gradient, and the norm regularization is taken to deconvolute the blurred image to recover the original image. As compared with the traditional methods, the proposed method estimates the obscure kernel of the image correctly, restrains the ringing phenomena well and improves the quality of remoter sensing image. The experiments for the same blade shows that Modulation Transfer Function(MTF) curve from proposed method is better than those from the blurred images and other reconstructed images.