本文提出一种新的结构字典学习方法,并利用它进行图像复原。首先给出结构字典学习的基本内容和方法, 然后将傅里叶正则化方法和结构字典学习方法有效整合到图像复原算法中。结构字典学习方法是先将原图像进行结构分解, 再分别学习出每个结构图像中的字典, 最后利用这些字典对原图像进行稀疏的表示。结合傅里叶正则化, 提出了一种有效的迭代图像复原算法: 第一步在傅里叶域利用正则化反卷积方法得到图像的初步估计; 第二步采用结构字典学习的方法对遗留的噪声进行去噪处理。实验结果表明, 提出的方法在改进信噪比和视觉质量上都要优于6种先进的图像复原方法,改进的信噪比平均提升0.5 dB以上。

In this paper, we propose a new structure dictionary learning method, and perform image restoration based on this approach. First, we define the structure dictionary for the nature image. Second, an iterative algorithm is proposed with the decouple of deblurring and denoising steps in the restoration process, which effectively integrates the Fourier regularization and structure dictionary learning technique into the deconvolution framework. Specifically, we propose an iterative algorithm. In the deblurring step, we involve a regularized inversion of the blur in Fourier domain. Then we remove the remained noise using the structure dictionary learning method in the denoising step. Experiment results show that this approach outperforms 6 state-of-the-art image deconvolution methods in terms of improvement signal to noise rate(ISNR) and visual quality, and the ISNR can be improved by more than 0.5 dB.