Image inpainting based on a non-linear partial differential equation

Abstract

Digital image inpainting is a process that attempts to automatically restore image. The objective is to automatically fill in missing data with simulated one, which has been achieved manually by professional painting restorators. The derivative information of image is very crucial in the inpainting since it contains the information of image structure. The structure tensor contains the information of gradient direction and tangential direction of an image. Hahn introduced a tensor regularizing model which preserves the image structure very well. Motivated by Hahn`s tensor regularizing model, we present a nonlinear PDE inpainting algorithm which uses the regularizing tensor. The regularized tensor in inpainting model is used as a coefficient of our anisotropic diffusion for image. Then it fills in the inpainting domain with the structure surrounding the inpainting domain. We provide some examples.