何凯, 牛俊慧, 沈成南, 卢雯霞
AuthorsHTML:何凯, 牛俊慧, 沈成南, 卢雯霞
AuthorsListE:He Kai, Niu Junhui, Shen Chengnan, Lu Wenxia
AuthorsHTMLE:He Kai, Niu Junhui, Shen Chengnan, 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:现有基于样本块的图像修复算法, 大多通过人工设定样本块大小来达到最佳修复效果, 缺乏自适应性; 此外, 对图像不同纹理和结构区域采用相同大小的样本块, 也不利于获得整体最优修复效果．为解决上述问题, 本文提出一种基于改进结构相似性的自适应样本块大小选取算法, 在传统的SSIM算法的基础上增加了梯度信息, 并通过结合样本块亮度、对比度和结构3个模块来衡量结构差异, 以此确定不同结构和纹理区域的最优样本块大小, 提高算法适应性, 改善修复效果．仿真实验结果表明, 当图像存在复杂的结构和纹理信息时, 本文算法仍然能够获得理想的修复效果．
Abstract_English:The current exemplar-based algorithms lack adaptability due to manually determining the size of block. In addition，using a patch with the constant size is not suitable to obtain the optimal effect in different structure and texture regions. To address this problem，this paper puts forward an adaptive patch size selection method using improved structure similarity(SSIM). The gradient information is added to the traditional SSIM and is combined with the brightness，contrast ratio，and structure of patch to measure the structural difference. On this basis，the optimal size of patch in different structure or texture regions is determined，thus improving the adaptability，as well as the inpainting effect. The simulation results demonstrate the effectiveness of the proposed method even when complex structure or texture exists.
Keyword_Chinese:图像修复; 纹理合成; 自适应样本块; SSIM算法; 梯度信息
Keywords_English:image inpainting; texture synthesis; adaptive patch; SSIM algorithm; gradient information

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6059