A Simulative Study of Analysis-By-Synthesis Perceptual Video Classification and Transmission over DiffServ IP Networks

This paper presents the results of transmission of video data on 2-class DiffServ IP networks using perceptual packet classification and slicing. An analysis-by-synthesis technique to identify perceptually important video regions, to create optimal video slices and to assign the resulting packets to the appropriate DiffServ classes is described. The proposed technique was implemented using the ISO/IEC MPEG-2 video coding standard. Several transmission scenarios, including homogeneous video traffic and interfering FTP traffic, were simulated using Network Simulator (NS). The proposed perception-based video transmission approach outperformed classical data partitioning in all tested network conditions, while delivering greater flexibility both in terms of network usage and potential to match time-varying channels. Substantially higher PSNR values than the regular best-effort case were also obtained assigning to the high-QoS class as little as 10% of the traffic. Demo sequences are available at http://multimedia.polito.it/icc2003.