【作者】 许林；

【导师】 史久根；

【作者基本信息】 合肥工业大学 ， 计算机系统结构， 2010， 硕士

【摘要】 H.264/AVC是国际电信联盟电信标准化部门ITU-T(ITU Telecommunication Standardization Sector)和移动图像专家组MPEG(Moving Picture Experts Group)联合成立的联合视频小组JVT(Joint Video Expert Team)于2003年3月正式发布的新一代视频编码国际标准,其目标是在保证图像质量的同时进一步提高数据压缩效率。H.264/AVC有着非常好的压缩性能,它将在数字电视广播、视频实时通信以及网络视频流媒体传递等领域发挥重要作用。论文首先在介绍视频编码技术的发展背景及现状的基础上,详细分析和讨论了H.264/AVC视频编码标准,并对H.264/AVC的特点、应用以及H.264编码标准中的帧内预测、帧间预测、整数变换、量化、熵编码和去方块滤波器等关键技术做了较为详细的分析和研究。其次,论文介绍了码率控制算法的基本原理。码率控制的目的就是控制输出码率,使输出码流更加平稳,从而更加有效地利用网络带宽,同时尽可能达到好的平稳的图像质量。论文介绍了RM8、TM5、VM8、TMN8等几种经典的码率控制算法以及基于拉格朗日优化算法的H.264编码控制模型,随后深入分析并讨论了H.264/AVC视频编码标准中使用的码率控制算法JVT-G012。JVT-G012码率控制算法主要由三步组成:目标比特分配;根据所分配的目标比特计算相应的量化参数;编码完成后对模型参数进行更新。论文重点研究了JVT-G012码率控制算法,并深入讨论了其存在的三点不足:一、JVT-G012码率控制算法中根据前一个GOP中所有P帧的量化参数均值来计算当前GOP中I帧的量化参数,对I帧的码率控制过于简单;二、JVT-G012算法仅根据缓冲区的剩余比特数对P帧进行比特分配,对P帧的比特分配不够精确;三、JVT-G012算法未考虑到视频序列中的场景切换问题,导致在场景切换处图像质量急剧下降,容易造成缓冲区溢出。针对上述三点不足之处,论文分别进行了处理:一、论文提出一种根据前一个GOP中I帧的复杂度比例关系计算I帧的量化参数,与JVT-G012算法中计算的I帧量化参数加权处理,对I帧进行较为精确的控制。二、根据输出缓冲区状态、P帧的图像复杂度以及P帧在GOP中的位置对P帧进行比特分配。三、针对JVT-G012算法未考虑到场景切换的缺陷,论文提出了一种基于场景切换的处理算法:根据像素点亮度分量绝对差值和均值法检测视频序列中是否发生场景切换;如果序列中有场景切换发生,使用论文提出的一种自适应GOP结构与长度的场景切换处理算法进行处理。最后,在JM8.6平台下,对JVT-G012算法与论文提出的算法分别进行实验验证。实验结果表明,改进后的算法能够更加精确控制输出码率,图像的PSNR(Peak Signal to Noise Ratio)也获得了一定的提高,同时改进后的算法能够有效的检测与处理视频序列中的场景切换,稳定了场景切换点后的图像质量。更多还原

【Abstract】 H.264/AVC is a new generation video coding compression standard. It was developed by JVT (Joint Video Expert Team) which was joined by ITU-T(ITU Telecommunication Standardization Sector) and MPEG(Moving Picture Experts Group) in March 2003. The goal of H.264/AVC is to ensure image quality and improve the compression efficiency. H.264/AVC has a very good compression performance. It will play an important role in the fields of digital TV broadcasting, real-time communication and network streaming meida.Firstly, the H.264/AVC coding standard was detailedly analyzed and discussed based on the introduction of the video encoding technology background and research status. The features and application of H.264/AVC and key techniques in H.264 which include intra-prediction, inter-prediction, integer transform, quantization, entropy coding and block filtering are detailedly analyzed.Secondly, the basic principle of rate control algorithm was introduced in this dissertation. The purpose of rate control is to control output bit rate, so that network bandwidth is used more efficiently and image quality will be better and more stable. We introduced several classic rate control algorithms such as RM8, TM5, VM8 and TMN8. H.264 coding control model based on the Lagrangian optimization algorithm was introduced, and the rate control algorithm of JVT-G012 used in H.264/AVC coding standard was deeply analyzed and discussed in the following. JVT-G012 rate control algorithm is composed mainly of three steps: target bit allocation; calculate the corresponding quantization parameter according to the allocated target bit; update the model parameters after the image was compressed.This dissertation focuses on the JVT-G012 rate control algorithm. There are three disadvantages of JVT-G012 algorithm have been discussed. (1). JVT-G012 rate control algorithm uses the P-frame’s quantization parameters of the previous GOP to calculate the I-frame’s quantization parameter in the current GOP, so rate control of the I-frame is too simple. (2). JVT-G012 algorithm is only based on the remaining number of bits of the buffer to allocate bits of P-frame, so the bit allocation of the P-frame was not sufficiently precise. (3). Because JVT-G012 algorithm does not take the problem of scene change in the video sequence into account, the image quality can decline sharply and the buffer can overflow easily when scene change occurred. To solve these problems, we dealt with them separately. (1). We calculate QP of the I-frame by weighting the QP which was calculated by using the proportion of I-frame’s complexity in the previous GOP and the QP which was calculated by H.264 rate control algorithm, to control the I-frame’s QP precisely. (2). We allocate bit of the P-frame based on the output buffer status, the image complexity of the P-frame and the position of the P-frame in the GOP. (3). Because JVT-G012 algorithm does not take the problem of scene change into account, we propose a rate control algorithm based on scene change in the dissertation. Firstly, we detect scene change in the sequence by using the mean absolute difference of pixel brightness component. If scene change was detected, we use an adaptive GOP structure and length of the scene change processing algorithm to process the scene change which was occurred.Finally, by using JM8.6 platform, the JVT-G012 algorithm and the proposed algorithm in the dissertation were carried out experimental verification respectively. Experimental results show that the improved algorithm can control the output bit rate more accurately, the PSNR of the image has also increased while the improved algorithm can effectively detect and deal with the scene change in the video sequences, and the image quality after the scene change has also been stabilized.更多还原