Abstract: Based on the sparse representation theory, this paper proposes an image inpainting framework by using different redundant dictionaries. First of all, it utilizes the Discrete Cosine Transform（DCT） or K-Singular Value Decomposition（K-SVD） to learn three different dictionaries, i.e., the redundant DCT dictionary, K-SVD global（KSVDG） dictionary and K-SVD Adaptive（KSVDA） dictionary. Then, the image to be inpainted is sparsely represented by using these three different redundant dictionaries, respectively. At last, based on the redundant dictionaries and sparse coefficients, the missing part of the image can be well expressed. Experimental results show that the proposed algorithm achieves good visual effects. Further, compared with several existing classical image inpainting methods, it produces superior results in terms of some main image quality evaluation indexes such as peak signal-to-noise ratio and feature similarity.

Key words: image inpainting, redundant dictionary, sparse representation

摘要： 在稀疏表示理论研究的基础上，提出了基于不同冗余字典的图像修补算法。首先设计采用离散余弦变换或K-SVD算法获得冗余DCT字典、KSVDG全局字典及KSVDA自适应字典等三种不同的字典；然后分别基于上述三种不同的冗余字典，稀疏表示待处理图像；最终图像中缺损的部分将通过冗余字典和稀疏系数有效地表示出来。实验结果表明，提出的算法修补后的图像视觉效果好，并在峰值信噪比、特征相似度等主要图像质量评价指标上优于现有几种经典的图像修补方法。

关键词: 图像修补, 冗余字典, 稀疏表示