基于稀疏表示和结构自相似性的单幅图像盲解卷积算法

发布时间：2018-06-24 14:39

本文选题：稀疏表示 + 结构自相似　；参考：《自动化学报》2017年11期

【摘要】：图像盲解卷积研究当模糊核未知时,如何从模糊图像复原出原始清晰图像.由于盲解卷积是一个欠定问题,现有的盲解卷积算法都直接或间接地利用各种先验知识.本文提出了一种结合稀疏表示与结构自相似性的单幅图像盲解卷积算法,该算法将图像的稀疏性先验和结构自相似性先验作为正则化约束加入到图像盲解卷积的目标函数中,并利用图像不同尺度间的结构自相似性,将观测模糊图像的降采样图像作为稀疏表示字典的训练样本,保证清晰图像在该字典下的稀疏性.最后利用交替求解的方式估计模糊核和清晰图像.模拟和真实数据上的实验表明本文算法能够准确估计模糊核,复原清晰的图像边缘,并具有很好的鲁棒性.
[Abstract]:Image Blind deconvolution Research on how to restore the original clear image from the blurred image when the fuzzy kernel is unknown. Due to the fact that blind deconvolution is an underdetermined problem, existing blind deconvolution algorithms make use of all kinds of prior knowledge directly or indirectly. A single image blind deconvolution algorithm combining sparse representation and structural self-similarity is proposed in this paper. The sparse prior and structural self-similarity priori are added to the objective function of image blind deconvolution as regularization constraints. Using the structural self-similarity between different scales of the image, the downsampling image of the observed blurred image is used as the training sample of the sparse representation dictionary to ensure the sparsity of the clear image in the dictionary. Finally, the fuzzy kernel and the clear image are estimated by alternating solution. The experiments on simulation and real data show that the proposed algorithm can accurately estimate the fuzzy kernel, restore the clear image edge, and have good robustness.
【作者单位】：清华大学电子工程系;北京工业大学计算机学院;
【基金】：国家自然科学基金(61501008) 首都卫生发展科研专项(2014-2-4025)资助~~
【分类号】：TP391.41

【相似文献】