Post-processing for intra coding through perceptual adversarial learning and progressive refinement

摘要

Video compression is an indispensable technology in image/video communications, due to its highly desirable ability to reduce the huge data volume. However, video compression introduces complex compression artifacts, such as blocking, ringing, and blurring artifacts. To reduce this issue, post-processing techniques have been extensively studied. In this paper, we propose a novel post-processing technique using multi-level progressive refinement network, which we term MPRNet. A joint training loss function is defined to optimize MPRNet via an adversarial training approach, which can boost both subjective and objective visual quality. Furthermore, our network employs progressive refine strategy, which predicts multi-level residues in one feed-forward pass. This way of coarse-to-fine refinement manner allows our network to make trade-off between refined quality and computational complexity, which facilitates the resource-aware applications. Extensive evaluations on benchmark datasets verify the superiority of our proposed MPRNet model over the latest state-of-the-art methods, achieves 8.2%, 8.3% and 9.5% of BD-rate savings for the Y/Cb/Cr channel over the HEVC baseline respectively. (C) 2019 Published by Elsevier B.V.