Image inpainting based on sparse representation using self-similar joint sparse coding

作     者：Zhang, Lei Chang, Minhui Chen, Rui 

作者机构：Yuncheng Univ Sch Math & Informat Technol Yuncheng 044000 Peoples R China Zhengzhou Univ Light Ind Software Engn Coll Zhengzhou 450001 Peoples R China 

出 版 物：《MULTIMEDIA TOOLS AND APPLICATIONS》 (多媒体工具和应用)

年 卷 期：2023年第82卷第13期

页      面：20215-20231页

核心收录：

学科分类：0808[工学-电气工程] 08[工学] 0835[工学-软件工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：National Natural Science Foundation of China Scientific and Technological Innovation Programs of Higher Education Institutions in Shanxi Province [2020 L0572] Scientific Research Project of Yuncheng University [XK-2018034, CY-2019025, YQ-2020021] Industrial Science and Technology Research Project of Henan Province [202102210387, 212102210418] Natural Science Foundation Project of Henan Province 

主　　题：Image inpainting Sparse representation Self-similar Joint sparse coding 

摘      要：In order to improve the sparse coding ability of over-complete dictionary and take advantage of the similarity between damaged pixels and their neighbors, we propose an inpainting method based on sparse representation using self-similar joint sparse coding. First, we perform singular value decomposition on the gradient vector of the image patches, and then divide the image patches into three categories: smooth patches, edge patches and texture patches according to the relationship between the primary direction and the secondary direction. Second, we use the KSVD method to train these three types of image patches respectively, and obtain three over-complete dictionaries that adapt to different local features. Third, we define a non-local self-similar matching function and use it to search for the most similar image patch to the current patch in the target region, and then use the similar patch and the current patch for joint sparse coding. Finally, we use the calculated sparse coding and the corresponding over-complete dictionary to reconstruct the current patch. A series of experimental results show that the self-similar joint sparse coding we proposed can not only improve the restoration effect of sparse representation methods to a certain extent, but also has good adaptability and can be combined with other sparse representation methods to improve their restoration effect.