An Image Inpainting Model for Grayvalue Images Based on Coherence Enhancing TV-H-1equation

The task of interpolating damaged or missing areas in an image using information from the neighborhood areas is known as Image Inpainting. Many PDE-based approaches are proposed in the literature for solving inpainting problems, but only some generate satisfactory results when solving the problem in a region containing coherent lines. In this paper we propose a fourth-order PDE for image inpainting using the TV-H
^-1
equation coupled with the Coherence Enhancing Structure Tensor to address the inpainting in the regions containing coherent structures. A numerical scheme is introduced based on splitting two convex energy terms to solve the proposed PDE model. The results suggest that our method joins curved isophotes in a better manner than generally accepted inpainting equations.