Content-adaptive robust error concealment for packet-lossy H.264 video streaming

摘要

In this paper, we present a spatio-temporal post-processing error concealment (EC) algorithm designed initially for a H.264 video-streaming scheme over packet-lossy networks. It aims at optimizing the subjective quality of the restored video under the constraints of low delay and computational complexity, which are critical to real-time applications and portable devices having limited resources. Specifically, it takes into consideration the physical property of motion field in order to achieve more meaningful perceptual video quality, in addition to the improved objective PSNR. Further, a simple bilinear spatial interpolation approach is combined with the improved boundary-match (B-M) based temporal EC approach according to texture and motion activity analysis. Finally, we propose a low complexity temporal EC method based on motion vector interpolation as a replacement of the B-M based approach in the scheme under low-computation requirement, or as a complement to further improve the scheme's performance in applications having enough computation resources. Extensive experiments demonstrated that the proposal features not only better reconstruction, objectively and subjectively, than JM benchmark, but also robustness to different video sequences.