Retrieving quality video across heterogeneous networks. Video over wireless

The article addresses the issues that arise in delivering video across Wireless Andrew. Wireless Andrew is the Carnegie Mellon University (CMU) heterogeneous networking environment. It interconnects wideband and narrowband wireless and wired networks using off-the-shelf technologies. The authors design a video system where video servers distributed across the network deliver, upon request, video clips to clients scattered around the campus. The main novelty of the article is generative video (GV), a content-based meta representation for video that is well suited for the heterogeneous dynamic environment of Wireless Andrew. GV provides a framework for graceful degradation as end-to-end network throughput varies. GV reduces the video sequence to a small set of still images and side information needed to reconstruct the original sequence. The authors have demonstrated for three real live video sequences that GV delivers compression ratios in the range of 1000-10000 with good perceptual quality They develop a scalable implementation for the GV coder/decoder (codec). Scalability coding avoids duplicating the encoding of each video clip when servicing a wide range of throughputs, as in heterogeneous networks. They discuss the video delivery requirements, the video system architecture, the system implementation, and the trade-offs needed to ensure graceful degradation under real-life network operating conditions.