可伸缩视频流的最优化决策及适配化传输

项目名称： 可伸缩视频流的最优化决策及适配化传输

项目编号： No.60802019

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 轻工业、手工业

项目作者： 邹君妮

作者单位： 上海大学

项目金额： 18万元

中文摘要： 可伸缩视频编码自2007 年完成国际标准H.264 扩展部分的统一制定，其多维分级性所表现出的信源率失真性能极大超越了传统分层编码，作为通信中的对偶问题，对可伸缩视频流进行信道优化的最大困难，在于若实现趋于网络信息理论边界的组播性能，就会损失其在信源率失真构造上的精度。本申请项目冀望保持可伸缩视频流的的信源-信道适配均衡，提出一种广义可伸缩流多维尺度渐进失真-率特性统计建模方法，采用基于Bayesian 网络有向图的概率推演，用于进行不同时间段纹理场景的多维尺度分级组合率失真预测；在此效用函数基础上，发展了两方面的调度和资源分配策略，建立对应马尔可夫决策过程和多径多层次有向子图空间模型，进行数学描述和规划求解。并且，引入网络编码的应用层组播来提升网络的吞吐性能。预期本项目的研究成果能应用于视频点播、IPTV、无线视频交互等领域。在掌握自主知识产权的同时，使我国在这一研究领域居国际前沿。

中文关键词： 可伸缩视频编码；分层组播；网络编码；马尔科夫决策过程；统计学习

英文摘要： As the expansion of H.264, scalable video coding (SVC) standard was finalized in 2007. Its excellent rate-distortion performance has greatly exceeded traditinal layered coding. To optimize the scalable video streams over the channel, the critical problem is to keep the tradeoff between the multicast capacity achieved by the nework information theory and the fine granularity of video coding. In this study, the adaptive equilibrium between the souce and the channel is discussed for the distribution of scalable video streams. A statistical modeling approach for progressive rate-distortion function is proposed. It uses Bayesian network-based probability inference to estimate combined rate-distortion of multi-dimension scalability in different texture scenario. Moreover, both scheduling and resource allocation are studied in the optimization target. Using convex programming methods, the Markov decision process and sub-graph model with multipath and multi-layer are well solvable. With the introduction of network coding, the overall throughput of the application-layer multicast can be effectively improved. The achievements of this study can be applied to many video applications,such as Video-on- Demand, IPTV and interactive wireless video. To grasp the intellectual property rights of these new techniques can increase our competence in this research area.

英文关键词： scalable vide coding; layered multicast; network coding; Markov decession process; statisical learning