透视场景的图像填补方法

摘要

针对户外场景中反映出的大量透视结构，提出一种基于交互的透视信息提取方法，并有效地应用于图像填补等问题上。首先通过用户协助获取交互接口，构建表征透视特征和结构特征的初始网格，并通过迭代过程修正不准确的用户输入；然后由几何计算生成完整的网格，从而令其尽可能准确地匹配透视方向和结构拓扑；最后，图像填补过程在透视网格的约束下进行，以保持源图像的透视结构。实验结果表明，该方法是简单、易用和友好的，而且成功地传递了透视信息，在被恢复的图像中保持了透视效果的一致性。%This paper presents an efficient interactive method for image editing and completion underperspective scenarios. By the method, an initial mesh representing perspective information and structure features was first obtained through user assistance, and corrected through iterations. Then an updated mesh that approaches the perspective direction and the structure topology was computed as accurately as possible from the initial mesh by geometry calculations. Finally, the filling was conducted under the perspective mesh in order to keep the perspec-tive effects of the original image. It can be seen from the experiments that this method is portable and easy-to-use, and as a result, the method can be used to extract 3D structure information with good visual consistency of per-spective effects in the recovered images.