Balanced interframe multiple description video compression

Abstract

A balanced twin-description interframe video coder is designed and performance results are presented for video transmission aver packet networks with packet losses. The coder is based on a predictive multiple description quantizer structure called mutually-refining DPCM (MR-DPCM). The novel feature of this predictive quantizer is that the decoder and encoder filter states trade in either of the two single-channel modes as well as in the two-channel mode. The performance and indeed the suitability of the multiple description approach for a network with packet lasses depends an the packetization method. Two packetization methods are considered-a "correct" one and a low latency but "incorrect" one. Performance results are presented for synthetic sources as well as for a video sequence under a variety of conditions.