A Real-Time N-Descriptions Video Coding Architecture

Abstract

In multiple description coding (MDC) literature two problems are still open: the capability of inserting any given amount of redundancy – in fact there is very often a minimum gap of redundancy under which it is impossible to go – and the opportunity to provide a high number of descriptions per frame – most of the architectures provide only two descriptions per frame. In this paper we focalize our attention on this second problem. A solution is to extend to N-descriptions the system given in [1], based on a polyphase down-sampler along rows and columns. Here multi-level scalability is introduced by generating the descriptions before the motion compensation loop. In this way there is no need of drift compensation terms and thus no structural constraints in scaling up the architecture.