Reducing Replication of Data in a Layered Video Transcoder

Abstract

In a previous work, we presented a Layered Video Transcoder (lvt) to transform an input H.261 stream in a set of data streams suitable for multi-rate layered multicast transmission. lvt performs a very simple strategy for filtering data, thus generates some amount of redundancy in the output streams. While this redundancy is useful to let receivers tolerate losses and change their subscription level without experiencing inconsistencies in the received video, reducing redundancy would grant a better compression rate. Furthermore, the performance analysis we performed suggested possible improvement to the lvt’s filtering algorithms. In this paper, we explore a set of new strategies to filter data, trading some additional complexity with better performance. In particular we discuss the possibility of doing a second conditional replenishment step in the transcoder and implement macroblock classification policies, using the history of macroblock updates to recognize the “role” of a macroblock in the video signal (e.g. background or moving subject). Performance analysis shows that these new strategies significantly improve the quality of video of the transcoder.