基于决策树的HEVC快速算法研究及系统实现
发布时间：2018-12-16 13:37
【摘要】：近年来伴随着2K、4K甚至8K高清视频的大量出现,针对高清视频内容的编码和传输显得更加重要,但是对于高分辨率的视频来说,传统的视频编码标准H.264/AVC的编码效率已经显得力不从心,因此为了提高编码效率,ITU-T的视频编码专家组VCEG和ISO/IEC的运动图像专家组MPEG联合成立了 JCT-VC组织,共同提出了最新一代的视频编码标准HEVC(High Efficiency Video Coding)。HEVC相比于H.264来说,改进的编码技术被应用于编码框架的各个模块当中,尤其是新的编码工具的提出使得HEVC的编码性能得到了很大的提升,但是在采用这些工具的同时,编码的时间复杂度也大大的增加,从而严重阻碍了 HEVC编码标准的推广和应用,所以本文针对HEVC的帧内编码优化和帧间编码优化进行研究,并利用符合HEVC编码标准的x265函数库进行多视点视频实时编码和传输的演示平台的搭建,本文的主要工作包括:(1)考虑到HEVC中基于率失真优化的编码单元四叉树递归划分算法存在着较大的时间复杂度,因此我们引入了一种利用特征数据进行离线训练的决策树预测模型,并将此预测模型运用在视频编码标准HEVC框架中,针对帧内编码单元(CU)的划分过程进行优化。决策树预测模型的引入有效的实现了帧内编码单元的快速划分;(2)在HEVC帧间编码的过程中,当前深度的编码单元需要遍历多种不同类型的预测单元(PU)模式,HEVC定义了 10种不同类型的PU划分模式,包含Skip/Merge模式、帧间2N×2N模式、帧间N×N模式、帧间N×2N模式、帧间2N×N模式等。我们依据实验统计的结果,利用F-Score选择算法进行特征的优化选择,将筛选的优化特征子集用于决策树预测模型的离线训练,从而提出一种依据决策树模型进行帧间预测模式的快速判决算法;(3)实现了多路视频实时编码和传输的演示平台的搭建。演示平台分为服务端和接收端两个部分,平台的界面基于MFC框架实现。服务端的核心主要依据FFmpeg编码视频方案,采用符合HEVC编码标准的x265函数库来进行视频的实时编码,采用实时流传输协议(RTP)和多线程编程技术将编码后的多路视频码流进行局域网传输;在接收端则依据FFmpeg解码视频方案,将接收到的网络数据包解码为视频数据之后,将其传递给视频显示线程并实时显示在屏幕的对应窗口。
[Abstract]:In recent years, with the emergence of 2K 4K or even 8K HD video, it is more important to encode and transmit high-definition video content, but for high-resolution video, The coding efficiency of H.264/AVC, the traditional video coding standard, has been inadequate, so in order to improve the coding efficiency, VCEG of ITU-T and MPEG of moving Image of ISO/IEC have jointly set up JCT-VC organization. In this paper, a new video coding standard, HEVC (High Efficiency Video Coding). HEVC, is proposed. Compared with H. 264, the improved coding technology is applied to all modules of the coding framework. Especially, the proposed new coding tools have greatly improved the performance of HEVC coding, but the time complexity of coding has been greatly increased while these tools are used, which seriously hinder the popularization and application of HEVC coding standards. Therefore, this paper studies the intra-frame coding optimization and inter-frame coding optimization of HEVC, and builds a demonstration platform for multi-view video real-time coding and transmission using x265 function library, which conforms to the HEVC coding standard. The main work of this paper is as follows: (1) considering the time complexity of quadtree recursive partition algorithm based on rate-distortion optimization in HEVC, So we introduce a decision tree prediction model which uses feature data for off-line training, and apply this prediction model to the video coding standard HEVC framework to optimize the division process of intra coding unit (CU). The introduction of decision tree prediction model effectively realizes the fast partition of intra coding units. (2) in the process of HEVC inter-frame coding, the coding unit of current depth needs to traverse many different types of prediction unit (PU) modes. HEVC defines 10 different types of PU partitioning modes, including Skip/Merge mode, inter-frame 2N 脳 2N mode, and interframe 2N 脳 2N mode. Inter-frame N 脳 N mode, inter-frame N 脳 2N mode, inter-frame 2N 脳 N mode, etc. Based on the experimental results, we use the F-Score selection algorithm to optimize the feature selection, and apply the optimized feature subset to the off-line training of the decision tree prediction model. Thus, a fast decision algorithm based on decision tree model is proposed for inter-frame prediction. (3) the demonstration platform of multi-channel real-time video coding and transmission is built. The platform is divided into two parts: server and receiver. The interface of the platform is implemented based on MFC framework. The core of the server is mainly based on the FFmpeg video coding scheme, and the x265 function library, which conforms to the HEVC coding standard, is used for real-time video coding. Real-time stream transmission protocol (RTP) and multithread programming technology are used to transmit the encoded multi-channel video stream to LAN. After the received network packet is decoded into video data according to the FFmpeg video decoding scheme, it is transferred to the video display thread and displayed in the corresponding window of the screen in real time.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN919.81
本文编号：2382439
[Abstract]:In recent years, with the emergence of 2K 4K or even 8K HD video, it is more important to encode and transmit high-definition video content, but for high-resolution video, The coding efficiency of H.264/AVC, the traditional video coding standard, has been inadequate, so in order to improve the coding efficiency, VCEG of ITU-T and MPEG of moving Image of ISO/IEC have jointly set up JCT-VC organization. In this paper, a new video coding standard, HEVC (High Efficiency Video Coding). HEVC, is proposed. Compared with H. 264, the improved coding technology is applied to all modules of the coding framework. Especially, the proposed new coding tools have greatly improved the performance of HEVC coding, but the time complexity of coding has been greatly increased while these tools are used, which seriously hinder the popularization and application of HEVC coding standards. Therefore, this paper studies the intra-frame coding optimization and inter-frame coding optimization of HEVC, and builds a demonstration platform for multi-view video real-time coding and transmission using x265 function library, which conforms to the HEVC coding standard. The main work of this paper is as follows: (1) considering the time complexity of quadtree recursive partition algorithm based on rate-distortion optimization in HEVC, So we introduce a decision tree prediction model which uses feature data for off-line training, and apply this prediction model to the video coding standard HEVC framework to optimize the division process of intra coding unit (CU). The introduction of decision tree prediction model effectively realizes the fast partition of intra coding units. (2) in the process of HEVC inter-frame coding, the coding unit of current depth needs to traverse many different types of prediction unit (PU) modes. HEVC defines 10 different types of PU partitioning modes, including Skip/Merge mode, inter-frame 2N 脳 2N mode, and interframe 2N 脳 2N mode. Inter-frame N 脳 N mode, inter-frame N 脳 2N mode, inter-frame 2N 脳 N mode, etc. Based on the experimental results, we use the F-Score selection algorithm to optimize the feature selection, and apply the optimized feature subset to the off-line training of the decision tree prediction model. Thus, a fast decision algorithm based on decision tree model is proposed for inter-frame prediction. (3) the demonstration platform of multi-channel real-time video coding and transmission is built. The platform is divided into two parts: server and receiver. The interface of the platform is implemented based on MFC framework. The core of the server is mainly based on the FFmpeg video coding scheme, and the x265 function library, which conforms to the HEVC coding standard, is used for real-time video coding. Real-time stream transmission protocol (RTP) and multithread programming technology are used to transmit the encoded multi-channel video stream to LAN. After the received network packet is decoded into video data according to the FFmpeg video decoding scheme, it is transferred to the video display thread and displayed in the corresponding window of the screen in real time.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN919.81
【参考文献】
相关期刊论文 前8条
1 苗健;许应康;;基于FFmpeg的视频播放软件设计[J];现代电子技术;2015年12期
2 胡聪;周甜;唐璐丹;;基于FFMPEG的跨平台视频编解码研究[J];武汉理工大学学报;2011年11期
3 杨红军;王金英;;基于VC++的多线程通信程序设计[J];电脑知识与技术;2008年29期
4 关晓蔷;梁吉业;钱宇华;刘煜伟;;基于决策支持度的决策树生成算法[J];计算机工程与应用;2008年27期
5 蒋志峰;;ffmpeg的快速音视频开发方法[J];单片机与嵌入式系统应用;2008年01期
6 单海涛;方向忠;;基于FFMpeg的高清数字电影软件编码系统的设计[J];信息技术;2007年01期
7 杨刚;;基于MFC用户界面设计主、子对话框数据的传递[J];机电产品开发与创新;2005年06期
8 袁媛;基于MFC类设计ObjectARX应用程序界面[J];中南大学学报(自然科学版);2004年03期
相关硕士学位论文 前1条
1 杨晓明;基于灰度统计特征的岩屑岩性最优描述方法研究[D];中国海洋大学;2007年
，本文编号：2382439
本文链接：https://www.wllwen.com/kejilunwen/xinxigongchenglunwen/2382439.html
