基于结构稀疏度和块差异度的目标移除图像修复

针对目标移除修复方法中存在的修复顺序不合理以及错误匹配问题,提出一种基于结构稀疏度和块差异度的图像修复方法。首先,在优先权中增加块的结构稀疏度计算,使优先权不仅依赖于目标块的几何特征,而且可以反映其邻域特征,提高了对目标块所处区域特征的辨识度,从而使修复顺序更加合理。其次,定义目标块与样本块之间的差异度,并在此基础上定义新的匹配规则,不仅对已存在像素之间的相似程度进行衡量,而且对已存在像素与填充像素之间的差异程度进行衡量,从而有效防止错误匹配以及错误累积。实验结果表明,该方法可以有效提高图像的修复效果,使修复图像更加符合视觉一致性要求。

Image Inpainting for Object Removal Based on Structure Sparsity and Patch Difference

Aiming at the problems of unreasonable filling order and the mismatch error in the image inpainting for object removal,an image inpainting method based on structure sparsity and patch difference was proposed.Firstly,the structure sparsity of the patch is added in the priority computation,because not only the priority depends on the geometric characteristics of target patch,but also its neighborhood characteristics are reflected,which can improve the identification of the regional characteristics of target patch,so that the filling order is more reasonable.Secondly,the difference between the target patch and the exemplar patch is defined,and the new matching rule is defined on the basis of this.In the new matching rule,it not only measures the similarity degree between existing pixels,but also measures the difference degree between existing pixels and filled pixels,thus effectively preventing mismatch error and error accumulation.Experimental results show that the proposed method can effectively improve the restoration effect,and make the restored images more consistent with the visual consistency requirements.