鲁棒的梯度驱动图像修复算法

数字图像形态特征的修复目前主要采用基于梯度驱动的偏微分方程(PDE)作为计算模型。虽然该类模型对较大区域的形态特征修复具有明显优势,但是修复过程中信息传播方向不确定使得它对修复对象具有选择性。在分析该类模型在图像修复中的计算本质和对应物理意义的基础上,结合典型仿真实验,认为保持信息传播方向始终指向待修复区域之外对修复结果具有决定性影响,并由此提出一种梯度驱动图像修复的新算法。实验结果表明,该算法能够保持信息传播方向的稳定,使得修复具有更强的鲁棒性。

Robust gradient driving image inpainting method

Gradient-driven PDEs (partial differential equations) are the main computing pattern for geometric inpainting models of digital images.Apparently,compared with previous models,gradient-driven computing models have a great advantage to the large-scale regions geometric inpainting,but its performances are not stable to different inpainted objects because the information propagating direction is uncertain in the inpainting process.Based on analyzing the computing essences and the corresponding physical meanings of gradient-driven models,it is decisive to the inpainting result that the information propagating direction always points to the outside of the inpainted regions.Thus,a new method of gradient-driven image inpainting is proposed.Experimental results prove that the method can stabilize the information propogating direction making its inpainting performance is more robust.