何凯, 高俊俏, 卢雯霞
AuthorsHTML:何凯, 高俊俏, 卢雯霞
AuthorsListE:He Kai, Gao Junqiao, Lu Wenxia
AuthorsHTMLE:He Kai, Gao Junqiao, Lu Wenxia
Unit:天津大学电气自动化与信息工程学院，天津 300072
Unit_EngLish:School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
Abstract_Chinese:传统基于纹理合成的图像修复算法, 其置信度值会随着修复过程的进行迅速衰减并趋近于0, 这会导致优先权计算结果不可靠, 产生错误的引导修复方向．除此之外, 传统方法通常采用最小绝对差平方和(SSD)准则来寻找最优匹配块, 匹配准则单一, 精度有限, 容易产生误匹配．为解决上述问题, 提出一种新的置信度更新函数, 以抑制置信度衰减过快的现象, 提高引导修复方向的准确性; 同时引入Census变换匹配准则, 将其与传统SSD匹配准则相结合, 以提高匹配精度．实验仿真结果表明, 本文算法鲁棒性较高且引导方向准确, 对于复杂的结构图像仍然能够获得理想的修复效果．
Abstract_English:In the traditional texture synthesis algorithms，the highest confidence value tends to reach zero rapidly with the process of inpainting，which will lead to an unreliable priority result as well as a misguided inpainting direction. In addition，the traditional methods usually search for optimal matching block using the sum of squared differences(SSD)criterion. In this way，the mismatches are inevitable because of the single matching criterion and the limited degree of precision. Therefore，a new confidence update function is proposed to avoid the rapid decay of confidence，which is helpful for improving the accuracy of guided inpainting direction. At the same time，matching accuracy is improved by combining Census transformation matching criterion with the traditional SSD. Experimental results show that the proposed algorithm is more robust，has higher accuracy in guiding direction and achieves a better effect even for the images with complex structure.
Keyword_Chinese:图像修复; 纹理合成; 置信度更新; SSD匹配准则
Keywords_English:image inpainting; texture synthesis; confidence update; SSD matching criterion

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=5812