| Traditional video coding techniques have been widely adopted in numerous products and services, such as TV broadcasting, video conferencing, mobile TV, and Blue-ray Disc. However, to support other application scenarios, for example, video delivery over heterogeneous networks and distributed video coding applications, distributed source coding, multiple path transmission and error resilience techniques are desired. Streaming video to diverse users over heterogeneous networks is a challenging problem, and it is critical that video applications work well over wired and multi-hop ad-hoc wireless networks. However, neither the network nor the application can provide these assurances working independently of each other. Motivated by these problems, we explore some key technologies that address these challenges.In this thesis, techniques for high compression efficiency and improved error resilience are proposed in several topics, including low bit rate video coding, distributed video coding, and joint source network coding. The contributions of this thesis are presented in Chapter 2 to Chapter 4.First, the theoretical foundations of wavelet based video coder as well as a new video coder for low bit rate video compression used in this research are presented. Our rate-distortion optimization (RDO) wavelet packet video coder, which codes the temporal subbands output from a motion-compensated temporal filter (MCTF), can determine optimal rate allocations in one GOP for a given coding bit budget. We propose three algorithms to find the best basis subtree of one temporal frame and optimal Lagrange multiplier of one GOP. The results show that the RDWP codec performance outperforms typically 0.6 to 0.9dB of the PSNR obtained by the MC-EZBC codec, through the entire range of low rates.Further, DVC tools for virtual correlation channel parameters estimation and improved Side Information (SI) are introduced for source classification based DVC system. After providing an overview of related topics, we provide theoretical analysis of how source classification can lead to improved rate-distortion performance. We perform pixel level online estimation of virtual channel model parameters at decoder, and propose a new SI generation scheme with iterative decoding witch uses a partially decoded frame. The improved SI results in a significant gain in reconstruction quality for video. The introduced tools improve the quality of decoded video, outperforming both the original codec and the DISCOVER codec.Then, a symbol-level network coding based wireless video conference system is described, where the listening scheme as well as error correcting scheme base on random linear network coding (RLNC) are presented. Network coding is attractive for a wide range of video applications. The goal of the listening is to reduce the number of transmissions or improve the coding gain. The RLNC based new mechanism can recover error of transmission symbols (LDPC codewords). The simulation results show that two improved schemes outperform the base system in terms of received video quality and coding gain by a significant margin.Finally, another joint source network coding application is studied. We address the scalable video bitstream protection and adaptation problem by a set of joint source network coding techniques for robust streaming video to heterogeneous users simultaneously. The bitstream is first divided into different priority substreams according to their effects to error propagation. Then unequal error protection is applied to all substreams. Symbols from the same priority level of several GOPs form one RLNC code. It is shown by simulation results that the proposed scheme distributes the impact of the loss of one GOP into partial data loss over multiple neighboring GOPs. The original video concent can be easily recovered with spatial/temporal error concealment in the proposed random linear network coding based scheme.
|
PDF Full Text Request