Nonlocal Curvature-Driven Diffusion Model for Image Inpainting

Abstract

A nonlocal image inpainting model is proposed by incorporating the nonlocal differential operators into the curvature-driven diffusion model. The new model differs from the original model in that pixels of similar structures rather than pixels in the neighborhood (the case for the original model) are utilized to estimate the lost pixels. This difference makes the new model performs very efficiently in inpainting images, especially textured images.