摘要： 利用单一字典修复包含多种结构成分的复杂图像的效果不太理想，针对该问题，提出一种基于三层稀疏表示的图像修复算法。该算法利用离散平稳小波、曲波和波原子稀疏来表示图像的光滑、边缘和纹理部分，采用块坐标松弛算法求解对应的稀疏优化问题实现图像修复。实验结果表明，该算法可以修复包含多种结构成分的图像，有效提高图像的修复质量。

关键词: 图像修复, 稀疏表示, 波原子, 曲波

Abstract: The quality of inpainting using a single dictionary is not so good that an image inpainting algorithm based on three-layer sparse representation is proposed. The algorithm makes use of discrete stationary wavelet, curvelet and wave atoms to implement sparse representation of smooth, edge and texture component respectively. Applying the block coordinate relaxation methods to solve the corresponding sparse optimization problem, the damaged image can be inpainted. Experimental results show that the algorithm can recover multiple structural components of the image and improve the quality effectively.

Key words: image inpainting, sparse representation, wave atoms, curvelet

中图分类号: 

• TN911.73