Adaptive bitstream switching of pre-encoded PFGS video

Abstract

With Progressive Fine Granularity Scalability (PFGS) video coding, one given encoding (with a prescribed bit rate) can flexibly be transmitted at any lower bit rate. However, the transmitted video is only efficiently encoded when the transmission bit rate is in the vicinity of the encoding bit rate; for transmission bit rates far from the encoding bit rate up on the order of 4 dB in video quality are lost. In this paper we develop and evaluate a suite of policies for accounting for this coding efficiency issue, which has been largely overlooked in previous PFGS streaming studies, in uni- and multicast streaming. Our adaptive policies select the PFGS encoding rate from a small number of pre-encoded versions and drop packets so as to maximize the reconstructed video qualities. Our policies consider both the visual video content, expressed using the motion activity level of MPEG-7 descriptor, as well as the channel variability. We find that an optimal non-adaptive streaming policy overcomes the 4 dB inefficiency and on top of this efficiency gain, our adaptive unicast streaming policy achieves 0.8 dB improvement over the optimal non-adaptive streaming. We also find that our content-dependent packet drop policies enforce fairness among multiple streams in terms of reconstructed video qualities and that our multicasting policy improves the average reconstructed video qualities at a group of receivers by up to 2 dB.