Transmission and quantization error concealment for JPEG color images using geometry-driven diffusion

In this article, we propose a new multi-resolution, geometry-driven error concealment method for missing data recovery on grey-level and color for any block-based compression schemes, such as JPEG or MPEG. It requires no extra redundancy bits nor source-destination interactions, as opposed to classical FEC/ARQ schemes. The reconstruction process consists in interpolating error-free decoded spatial information into corrupted areas, using the actual geometry of local structures. In addition, a multi-resolution scheme is used to organize the recovery process, from coarse to fine structures, and achieve better performances in terms of approximation and computation time. We demonstrate that beside transmission errors, our model can also be used to conceal quantization errors (compression artifacts).