Inpainting for compressed images

Abstract

Motivated by the recent work on image inpainting in pixel domain using tight frame in J.-F. Cai et al. (2008) [3], we propose a unified iterative frame-based algorithm for inpainting compressed images to recover missing bits or missing coefficients during the compression process. Such an inpainting algorithm could be applied to any image compression scheme based on coefficients thresholding and quantization in some transform domain as a post-process to effectively remove compression artifacts, or equivalently to improve the compression ratio. The convergence of the iteration is proved and the resulted solution minimizes a special functional. Numerical experiments on popular image compression schemes demonstrate the effectiveness of our inpainting algorithm on improving the visual quality of compressed images. © 2010 Elsevier Inc. All rights reserved.