项目名称: 面向众核处理器的高并行度视频编码关键技术研究
项目编号: No.61272323
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 张勇东
作者单位: 中国科学院计算技术研究所
项目金额: 80万元
中文摘要: 众核处理器为视频编码提供了强大的计算能力,也给并行视频编码研究带来了极大的挑战。鉴于目前的并行视频编码方法并行度低,无法充分利用众核处理器的并行计算能力,本项目拟开展高并行度视频编码的关键技术研究。针对环路滤波模块控制密集的问题,首次引入任务级并行,并针对两个任务存在的问题,拟分别采用马尔科夫转移概率加速方法和独立像素连通区域并行方法予以解决;针对熵编码模块数据相关性强的问题,在熵编码片分割上拟限制熵编码片的大小,更改熵编码片的组织方式和初试上下文模型,在语法元素分割上对语法元素集合采用流水线并行;针对HEVC并行运动估计存在的问题,拟提出一种PU级的全局并行方法,用新颖的边界预测块处理顺序提高并行度,复用已知信息预测运动矢量以保证编码效率;最后,构建基于众核处理器的并行HEVC视频编码原型系统,对研究成果进行集成化的验证和展示,以推动面向众核处理器的高并行度视频编码技术的发展。
中文关键词: 众核处理器;环路滤波;帧内预测;运动估计;并行处理
英文摘要: Many-core processor provides powerful computing ability for video coding, but also brings a great challenge to the research on parallel video coding. Existing parallel video coding methods have low degree of parallelism, which can't make full use of the computing power of many-core processor. In the project,we will carry out the research on key techniques for highly parallel video coding on many-core processor, including three key modules, and a prototype system. The control of deblocking filter module is very intensive, and so we will be the first to introduce the task-level parallelization. According to the problem of two tasks, we intend to use Markov transition probability method, independent pixel connected region method, respectively. Because the entropy coding module has strong data dependency, we intend to limit the size of the entropy coding slice and change the initial context model and the organization of entropy coding slice, meanwhile we will use pipelined parallel method for different sets of syntax elements. According to the problem of parallel motion estimation for HEVC,we will develop a novel global parallel method based on PU-level for motion estimation, which includes a novel coding sequence for border prediction unit to improve the degree of parallelism and a reuse mechanism of temporay infor
英文关键词: Many-Core Processor;Deblocking Filter;Intra Prediction;Motion Estimation;Parallel Processing