特征聚类的局部敏感稀疏图像修复

针对图像修复过程中单一的字典迭代时间长、适应性差、修复效果不理想的缺点，提出了一种结合图像特征聚类与字典学习的改进的图像修复方式。首先破损的图像被分割成小块，并产生索引矩阵。然后使用控制核回归权值算法，对其进行图像聚类。通过对图像内在结构与未破损区域信息的挖掘，分割的图像块根据SKRW的相似性进行了分类。之后针对不同类型结构的图像，通过自适应局部明感字典学习的方式，获取每类字典的过完备字典。然后，通过构建自适应局部配适器，提高字典更新的收敛速度与稀疏字典的适应性。因为是通过多个字典匹配不同结构的图像，因此图像的稀疏表示更为准确。各个字典在达到收敛之前不断进行更新，而图像的稀疏因子也会随着改变。在对破损区域进行补丁更换之后，实现了对破损图像的修复。实验结果表明，该算法相较于目前的修复算法，视觉效果和客观评价上更好，且所需的修复时间更短。

Image inpainting based on feature clustering and locality-sensitive sparse representation

A novel image inpainting method based on sparse representation which combined image clustering and dictionary learning was proposed to solve the problems of long iteration time, bad adaptation and non-ideal results when using one single dictionary. Firstly, the broken image was divided into blocks and generated index matrix. Then Steering Kernel Regression Weight (SKRW) algorithm was used for image clustering. By exploring the inner structures of image and the information of intact area, blocks were sorted into categories based on their similarities of SKRW. Then each category had their own overcomplete dictionary by self-adaptive locality-sensitive dictionary learning. By building a self-adaptive local adaptor, the rate of convergence and the adaptability of sparse dictionary were improved. Multi-dictionaries were matched with different image structures, so the image would have a more accurate sparse representation. The dictionaries were updated until convergence, along with sparse coefficients as well. The image was finally restored after replacing patches back. Experimental results show that the proposed algorithm can repair the damaged images better than the state-of-the-art algorithms in both visual effect and objective evaluations. In addition, the time consumption is greatly reduced in comparison with the other algorithms.