【作者】 郑莉华；

【导师】 吴跃；

【作者基本信息】 电子科技大学 ， 计算机软件与理论， 2013， 博士

【摘要】 由ITU和MPEG联合推出的H.264/AVC编码标准是目前应用最为广泛的视频编码标准之一，该标准在网络视频服务等实时应用领域中正逐步取代原有视频压缩标准占据主导地位。H.264/AVC采用的码率控制技术在为用户提供较高的压缩比和良好的视频重构质量的同时，也使H.264/AVC的计算复杂度更高，编码同样的视频序列需要耗费更多的时间，这对低延迟的实时视频应用来说是非常不利的。另外，高计算复杂度的视频编码过程也对处理器的计算能力提出了更高的要求，并不是所有的处理器都能满足视频编码的实时需求。将视频服务系统构建到云计算平台上，并行地编码视频序列是解决上述问题的一个有效方法。本论文在对H.264编码器及云计算技术进行系统、全面地分析与总结的基础上，深入细致地研究视频服务系统中编码失真和编码速度涉及的关键技术，主要工作包括：1.针对H.264/AVC采用的二次模型及线性跟踪模型存在的预测值误差较大的问题，提出一种新的联合率失真模型。分析表明新的联合率失真模型能很好地体现R与QP和C之间的关系。2.针对H.264/AVC在编码模式选择上的高计算复杂度，根据I帧和P帧不同特性，分别提出帧内编码复杂度计算模型和帧间编码复杂度预测模型。帧内编码复杂度计算模型用几何梯度作为复杂度的衡量方法，有效降低了编码的计算复杂度，加速了帧内编码过程。帧间编码复杂度预测模型通过四种预测方法及预测方法的优选，有效降低了帧编码复杂度的预测误差。3.针对H.264/AVC在视频中包含快速运动或场景突然切换时量化参数计算不准确、码率失配的缺陷，提出一种全I帧码率控制算法和一种全P帧码率控制算法。通过在量化参数计算中引入当前I帧的编码复杂度，以及对I帧和P帧采用不同的位分配策略，新的码率控制算法很好地弥补了H.264/AVC在上述两方面的不足。实验表明，新的码率控制算法较之于JVT-W042可以获得更高的峰值信噪比增益，缓冲区充盈度控制更准确，跳帧次数更少，预测的帧编码复杂度更精确，误差更小。4.针对现有视频服务系统在低延迟的实时应用领域存在的资源利用不均衡和处理大规模数据时的低效率，提出一种基于Map/Reduce的并行视频编码架构。论文根据视频编码的异构性和动态性特点，将云计算的虚拟技术及任务调度策略应用于视频服务。实验表明该视频并行编码架构能有效地提高视频编码速度。5.针对视频编码帧特性及相应的编码复杂度计算模型，提出云环境下的最大-最小负载均衡（Load Balance Max-Min,LBMM）任务调度算法。该调度算法综合考虑视频子序列的帧编码复杂度及各服务器节点的实时状态，优化视频子序列的调度分配。实验结果表明，LBMM调度算法可大幅降低整个视频编码的时间跨度。更多还原

【Abstract】 The H.264/AVC encoding standard jointly launched by ITU and MPEG is one ofthe most widely used video encoding standards and it replaces the existing videocompression standard in the real-time applications dominant such as network videoservices.The rate control technology adopted by H.264/AVC can improve the videocompression rate and provide users with good video reconstruction quality, but theyincrease the H.264/AVC computing complexity. So it will take more time-consuming toencode the same video sequence. This is very unfavorable for low-latency real-timevideo applications. In addition, the video encoding is a complicated calculation processand it needs very high processing capability of the computer. Currently, the computecapability of a single processor can’t meet the demand for real-time application in thevideo encoding.It is an effective method to solve the above problems by constructingthe video service system onto the cloud computing platform and parallel encoding thevideo sequences.Based on a systematical summary of relevant works on H.264encoder and cloudcomputing,this dissertation thorough research on the key technologies of codingdistortion and coding speed involved in the video service system.The main worksinclude:1. A noval joint rate-distortion model based on the exponential function is proposed.The model aims to resolve the larger prediction error problem in quadratic model andlinear tracking models which are adopted by H.264/AVC. The analysis shows that thenew joint rate-distortion model can well reflect the relationship between R,QP and C.2. According to the different characteristics of the I-frames and P-frames, theintraframe encoding complexity calculation model and interframe encoding complexityprediction model are proposed,aiming at the high computational complexity of theH.264/AVC encoding mode selection. The intraframe encoding complexity calculationmodel utilizes the geometric gradient as a measure of the complexity and it caneffectively reduce the computation complexity of the encoding to accelerate the intra-frame encoding process. The interframe encoding complexity prediction modeluses the best one of the four forecasting methods to effectively reduce the frameencoding complexity prediction error.3. In order to resolve these problems that quantization parameter calculation is notaccurate and the bit-rate mismatches in fast motion or scene suddenly switch forH.264/AVC video, a intra-only rate control algorithm and a P-frame rate controlalgorithm.are proposed. Introducing the current I-frame encoding complexity to thequantization parameter calculation and the I-frames and P-frames with different bitallocation strategy, the new rate control algorithm compensates H.264/AVC inadequatein the above two aspects. The experiments show that the new rate control algorithm canachieve higher PSNR gain and the buffer filling degree controls more accurately, frameskipping fewer, more precisely predicted frame encoding complexity and the errorsmaller compared to JVT-W042.4. A parallel video encoding architecture based on the Map/Reduce is proposed.Inorder to resolve the problem that resource utilization is unbalanced and large-scale dataprocessing is low efficiency for video service system in the real-time applications withlow latency,this dissertation adopts virtualization technology and task schedulingstrategy in cloud computing according to the the heterogeneous and dynamiccharacteristics of the video encoding.The experiments show the video parallel encodingsystem can effectively improve the speed of video encoding.5. A Load Balance Max-Min(LBMM) task scheduling algorithm under a cloudenvironment based on video encoding frame characteristics and the correspondingencoding complexity computing model.is proposed. The scheduling algorithm considersthe real-time status of the video sub-frame encoding sequence complexity and servernodes to optimize the allocation of video subsequence scheduling. The experimentalresults show LBMM scheduling algorithm can significantly reduce the entire time spanof the video encoding.更多还原