【作者】 魏振宇；

【导师】 张旭东；

【作者基本信息】 清华大学 ， 信息与通信工程， 2005， 硕士

【摘要】 从视频电话,视频会议到 DVD,数字电视,机顶盒,数字视频在我们的周围无处不在,深深的丰富了我们的生活。针对数字视频的不同应用领域,国际组织提出了一系列视频压缩国际标准。正是由于众多视频标准的发展使数字视频得到广泛的应用。同样,数字视频应用领域的扩大,也为视频压缩技术提出了更高的要求,推动了视频压缩标准的发展。H.264/AVC 是目前由 ISO MPEG 和 ITU-T VCEG 联合提出的最新一代视频压缩国际标准,和以往视频编码标准相比,H.264 采用了很多特殊的技术以降低码率,提高性能,例如:1/4 精度的亚象素运动矢量搜索、多帧预测、7 种模式的块匹配、以及帧内预测、整数 DCT 变换、CABAC 或 CAVLC 的熵编码技术等等。这些技术的使用,使得 H.264 具有低码率,高质量的性能。但是这些技术的使用也大大增加了编码器的复杂度,使得运算量大大增加,远远不能满足实时编解码的要求。如何对 H.264 的算法进行优化,使其满足实际工程需要成为近1～2 年来研究热点。通过对算法的研究,作者发现运动估计,亚象素搜索以及模式选择部分耗费了编码器运算量的 70%～80%,对其进行优化改进很有必要。在充分利用图像相邻宏块间具有较强空域相关性的基础上,本文提出了基于预测和早停止技术的快速运动估计算法和亚象素搜索算法,以及基于模式预测和自适应阈值的快速模式选择算法,显著的提高了编码器的运行效率,并较好的保持了图像的质量,对视频压缩码率也没有明显影响。随着数字信号处理器(DSP)的快速发展,为实现便携式的数字视频通讯终端提供了可能。TI(Texas Instruments)公司推出的 C64 系列 DSP 具有高主频,并行处理能力高的特点,是目前处理能力最高的数字信号处理器之一。将 H.264 的算法实现在以 C64 DSP 为核心处理器的系统中,具有较强的工程意义和市场价值。本文将 H.264 的编码器实现在 Ti C6416DSP 的仿真开发平台 NVDK 中,通过系统级优化,算法级优化,指令级优化等手段,充分利用 DSP 的并行处理能力和丰富的媒体处理指令,最终达到了实时编码的效果。更多还原

【Abstract】 Digital video is being adopted in an increasingly fields of applications rangingfrom video telephony and videoconferencing to DVD, digital TV and top-set box. Itenriches our living deeply. Aiming at different application fields of digital video,international organization recommended a serials of international video compressionstandards. The adoption of digital video in many applications is due to thedevelopment of many video coding standards。In the same way, extension of thedigital video application fields produces more requirements for video compressiontechnology and pushes the development of the video coding standards, too.H.264/AVC is a new recommendation for moving picture coding proposed byITU-T VCEG and the ISO/IEC Moving Picture Experts Group, in which a lot of newtechniques are used for improving encoding efficiency and reducing the bit-rate., suchas 1/4-pel fractional pixel motion estimation, multi-frame motion estimation, multipleblock size, intra prediction, integral DCT transformation, CABAC and CAVLCentropy coding etc. With the use of these techniques, encoding quality is improved,but the complexity of the encoder and computational cost are increased at the sametime. Optimizing the algorithms in H.264 and making them suitable for practicalapplications has been a hot research field in recent years. By studying the algorithmof H.264, it can be found that motion estimation, sub-pixel search and mode selectionconsume about 70%~80% computational cost of the H.264 encoder, so it is necessaryto optimize these parts. Based on highly correlation among the adjacent Macroblocks,a fast motion estimation algorithm and a fast sub-pixel search method are proposedby using prediction and early termination techniques. A fast mode decision algorithmis also proposed by using mode prediction and adaptive threshold techniques. Thesemethods could increase the efficiency of the encoder and keep image quality quitewell with almost the same bit rate.With the fast development of the digital signal processor (DSP), it is feasible toachieve portable digital video communication terminal. C64 family DSPs producedby TI Corp.(Texas Instruments) have the features of high frequency and highparallelism. It is the most powerful digital signal processor at present. It has greatengineering significance and business value to implement H.264 codec in the C64DSP system. We choose ATEME NVDK as our developing platform, on which wecomplete H.264 codec. By using the way of system optimization, algorithmoptimization and instruction optimization, we have successfully implemented theH.264 encoder on NVDK and made great improvement in the aspects of codingperformance and real-time completion.更多还原