提出一种曲率驱动的方向性热流图像修复模型。在该模型中,信息向待修复区域的扩散被看成是两个相互垂直的1D方向性热流的耦合,其中的扩散方向由图像的局部相关几何决定,而扩散强度是一个曲率相关的函数。实验结果表明,该模型不仅能够保证修复边缘的尖锐性还可以减少修复过程中虚假边缘的产生。
An inpainting model which based on curvature driven oriented heat-flows is proposed in this paper. In this model, the diffusion toward the inpainting domain can be seen as the coupling of two weighted 1 D heat flows along orthonormal directions. The diffusion directions are defined by the local coherence geometry of image and the diffusion strength is a function of curvature. Experimental results show that the emergence of artificial edges can be avoided and the sharp inpainting edges can be achieved by using this model.