偏微分方程约束下基于动态大小模块的图像修复模型

A PDE Constraint Adaptive-Exemplar Based Image Inpainting Model

提出一种保持图像线性特征和纹理信息、并能应用于存在较大规模缺失信息区域的图像修复模型.该模型根据图像局部纹理特征强度动态地选择匹配模块的大小,使用切向等照度线强度数据项约束修复顺序.几何特征作用下的切向等照度线偏微分方程在边缘宽度约束下扩散,具有很好的线性特征保持性能;它具有形态学不变性,可以真实地修复自然场景图像.根据基于模块的修复模型的特点,使用欧氏距离作为模块相似度测量函数.针对模型引起的模块接缝效应,使用散度约束下总体变分插值法对修复结果进行无接缝效应处理,使得最终结果平滑.理论和实验证明了该模型在图像修复中的有效性.

A novel image inpainting model based on large target region exemplar is proposed. The model can preserve the linear structure and texture information of the image, and adaptively select exemplar based on local texture information. The model uses cross isophotes data term to constrain inpainting order. Cross isophotes data term is the result of partial differential equation （PDE） under the control of image geometry property. By taking into consideration of the extent of edge, the model shows a good linear structure preserving property. Diffusion of cross isophotes is morphologically invariant in the method, and the target region of natural scene could be more plausibly filled up. According to the feature of exemplar-based inpainting, Euclidean distance is used as similarity measure. To reduce seams caused by this model, a gradient constrained total variation （TV） interpolation is used in the inpainting result to obtain a seamless image. Both theoretical analysis and experiments verified the validity of the new inpainting model.