Abstract
In recent years, video transmission is rapidly developed and widely employed in various applications; however, no matter the traditional wired network or the latest wireless network, they are all evolving to packet-based but currently provide no or limited end-to-end QoS guarantees. Packet loss is unavoidable when real-time video is to be transmitted over the networks. In order to overcome the problem, based on the concept of multiple descriptions (MD) coding, we propose a new MD coder: Multiple Description Bit-plane Coding (MDBC) which combines the alternative and duplicate schemes to produce two descriptions to be transmitted over networks.From the experimental results, when the bit rate is 144kbps, our MDBC coder outperforms conventional coders in average by 5.54 dB, 6.75dB and 2.77 dB for Akiyo, Foreman, and Stefan sequences, respectively. Besides, for comparison with different MD coders, we also present the performance of MDSQ. At packet loss rate of 10% and 20%, our MDBC provides about 0.23dB, 0.23dB, and 0.38dB in average better performance than MDSQ for Akiyo, Foreman and Stefan sequences, respectively. However, for the same sequences, our MDBC has 1.04dB, 0.34dB and 0.14dB loss in low packet loss cases.For the packetization in common MD coders, we propose Interleaved Packetization which utilizes two different scan orders to packet MBs into video packets of two descriptions. Therefore, the probability that descriptions of adjacent MBs are lost simultaneously is reduced significantly so that the performance of the error concealment can be improved. In our experiment, on the average for all packet losses, our interleaved packetization is superior to raster scan by more than 0.44 dB PSNR in average.