An efficient framework for image/video inpainting

Abstract

Image inpainting has been widely applied to many applications, such as restoring corrupted old photos, erasing video logos, concealing errors in a digital video processing system, and so on. However, traditional geometric inpainting methods suffer low efficiency. To tackle this problem, this paper addresses an efficient transform based framework for geometric methods. Given an image, we firstly decompose it, then separately perform restoration process and finally employ Laplacian diffusion function to hold local texture coherence. Experimental results show that the proposed method not only speeds up and enhances the performances of geometric methods, but also obtains a better restoration results compared with the traditional texture and hybrid methods.