块关联匹配与低秩矩阵超分辨融合的图像修复.pdfVIP

第27 卷 第2 期 计算机辅助设计与图形学学报 Vol. 27 No.2 2015 年2 月 Journal of Computer-Aided Design Computer Graphics Feb. 2015 块关联匹配与低秩矩阵超分辨融合的图像修复 马 爽, 谈元鹏, 许 刚 (华北电力大学电气与电子工程学院 北京 102206) (lanxi1279@126.com) 摘 要: 针对块匹配图像修复算法容易出现结构不连贯与块效应问题, 提出一种两阶段图像修复方法. 该方法通过 增加相邻修复块在重叠区域的关联性约束, 利用图像块关联匹配修复算法实现对降采样受损图像的粗修复, 以保证 图像主体结构的完整性; 依据不同分辨率图像中像素点与其 8 邻域的线性相关性, 引入带噪低秩矩阵填充算法对图 像细节进行超分辨率精修复, 使修复后图像的纹理与色彩变化具有连续性. 实验结果表明, 文中方法对包含渐变特 征的自然图像修复具有更好的视觉效果. 关键词：图像修复; 块关联匹配; 低秩矩阵; 超分辨率 中图法分类号：TP391 Image Completion Based on Fusion of Patch Associated Matching and Low-rank Matrix Super Resolution Ma Shuang, Tan Yuanpeng, and Xu Gang (School of Electrical and Electronic Engineering, North China Electric Power University, Beijing 102206) Abstract: A novel two stage method of image completion is proposed to handle structure discontinuities and block artifacts caused by traditional exemplar-based image completion algorithms. With the relevance of adjacent patches taken into consideration, a patch associated matching restoration algorithm, which is able to fix main structure in image, is performed on the down-sampling image. Meanwhile, according to the linear correlation of pixels and their eight neighbors in different resolution images, a super-resolution method based on noisy low-rank matrix filling is proposed and employed to refine the target image. This operation ensures the continuity of the changes in the detail information of texture and color. Experimental results demonstrate that the proposed method has more outstanding performance on natural images with gradual features. Key words: image completion; patch associated matching; low-rank