Nonlinear anisotropic diffusion-based structural inpainting framework

Abstract

We consider a novel structure-based image inpainting technique in this paper. The proposed interpolation approach is based on a nonlinear parabolic partial differential equation (PDE) model based on anisotropic diffusion. A novel edge-stopping function is properly modeled for this diffusion-based reconstruction scheme that also uses an inpainting mask. The developed differential model is well-posed, admitting a unique and weak solution that is computed by using a robust finite difference-based numerical approximation scheme. Successful interpolation experiments and method comparison are also described.