Abstract: Aiming at priori knowledge representation of image inpainting algorithm based on sparse representation, considering image texture self-similar and group-structure spasity of atom’s coefficient, a group-structured sparse representation model is proposed. In the model, the sparse coefficient of adjacent atoms is restrained by appropriate group structure, and the input image valid patches and training samples are unified for joint sparse representation and learning dictionary, in which each element of the patches has the same sparse pattern, then this relationship is used as priori knowledge for image inpainting. Experimental results on target removing and pixels lost inpainting show that the proposed method has good performance.

Key words: information processing technology, sparse representation, dictionary learning, group-structured

摘要： 针对基于稀疏表示的图像修复方法存在稀疏系数先验知识表达不足等问题，考虑图像的纹理自相似性和原子系数的群结构稀疏性，提出了群结构约束的稀疏表示模型，通过选取合适的群结构约束稀疏系数，使字典中相邻基对应的稀疏系数之间建立联系，并统一对输入图像的有效数据图块与训练样本进行稀疏编码来进一步训练字典，使其具有相同的稀疏模式，从而建立联合稀疏关联，并将其作为先验知识指导图像修复。通过区域目标剔除、像素缺失修复等实验验证其性能，实验结果表明，该方法有较强的自适应性，修复效果较好。

关键词: 信息处理技术, 稀疏表示, 联合字典学习, 群结构