延迟约束的视频编码并行优化研究

With the rapid development of high QoE video services like Ultra HD and VR panoramic video broadcasting, video coding need to address the following challenges: the capability of “eating” big video data, the capability of encoding fast enough and the capability of encoding with high efficiency. Today, parallel video coding solution based on generic heterogonous parallel computing platform has been a necessity. But the coding performance degrades by the broken of prediction dependence for such parallel realization. In this research, instead of evaluating the algorithms performance by assuming serial realization, we evaluate the complexity and compression performance of the algorithms from the perspective of parallel realization. Based on a general platform with multi-or many-cores processors, we conduct a research on the low delay, parallelized video coding optimization technology by focusing on both parallel coding and low delay. We propose a novel framework for parallel complexity analysis and parallelized complexity-rate-distortion joint optimized model. Moreover, we propose a rate distortion compensation method for parallelized video coding, an optimized resource allocation method in context of parallel computational resources and a low delay and parallelized rate control approach to optimize video coding in terms of efficiency, delay and processing capability. The expected achievements from this research will break the technical bottle-neck of the high quality video coding and bring a high performance encoder based on generic computing platform and will have a significant impact on both academy and industry.

随着超高清以及VR全景视频直播等高用户体验服务的迅猛发展，视频编码面临着“编得下”、“编得快”和“编得好”三方面的挑战。基于通用异构计算平台的视频并行编码方案成为应对上述挑战的必然途径。但并行化的视频编码必然打破预测编码的前后依赖性，降低率失真效率。本项目打破传统的以串行化实现为依据的算法设计和理论研究，针对主流的异构多处理器计算平台，强调编码的并行化和低延迟，研究低延迟并行化视频编码优化技术。通过建立并行复杂度分析框架和并行化的复杂度-码率-失真联合优化模型，提出视频并行编码的率失真优化补偿方法、并行计算资源优化分配方法、低延迟码率控制方法等多项创新技术，我们从效率、延迟和处理能力三个维度对视频编码优化问题进行了全面创新研究。研究成果将突破高品质视频服务面临的技术瓶颈，形成基于通用计算平台的高性能视频编码方案，具有重大研究价值和产业推动作用。

编码端复杂度和延迟，传输端码率，用户端感知视频质量三方面的权衡，成为视频编码应用中的关键问题。本项目首先分别从编码端、传输端和用户端“三端”进行方案设计和独立优化。为优化编码复杂度并实现低延迟，本项目以并行异构场景中的编码速度优化为核心，提出并行化和低延迟场景下的视频编码复杂度分析、优化、计算资源分配等多项创新技术。为实现传输端低码率，设计了以负载均衡为依据的率失真性能的并行化补偿策略，进而形成并行码率控制、特定场景编码传输方案。为准确评估并优化用户端视频质量，设计了逼近人眼视觉系统特性的质量评测方法，提出了系列视频预处理、后处理技术，基于此形成创新性编码-传输主观视觉优化方案。通过建立并行化的复杂度-码率-失真模型，对“三端”进行联合优化并设计和构筑了一整套包含视频编转码、视频处理功能的分布式媒体处理系统。该媒体处理系统从压缩率、延迟和处理能力三个维度对视频编码优化进行了有机的结合。整体上，形成了基于通用计算平台的高性能视频编码方案，实现了现有主流编码器所不具备的编码速度-性能-延迟的权衡以及不同应用场景的可拓展性，具有重要的产业潜在应用价值。共发表学术论文56篇，其中SCI期刊4篇，会议论文47篇，做特邀技术报告14次，申请发明专利23项，其中授权9项。已培养毕业硕士研究生22名，在读博士研究生5名。