【作者】 张帆；

【导师】 张旭东；

【作者基本信息】 清华大学 ， 信息与通信系统， 2004， 硕士

【摘要】 H.264是ITU-T VCEG和ISO MPEG联合开发的最新视频标准,它的高编码性能引起了人们的广泛关注。然而,H.264的高运算量使其在现阶段还很难得到广泛的应用。本文的工作主要着重于对H.264的编码算法进行算法级与指令级的优化。运动估计是视频编码器中最重要的模块,其算法直接影响到编码的性能与速度。而H.264第一次引入了7种宏块划分模式,并将运动补偿的精度提高到了1/4像素,进一步增大了运动估计的运量复杂度。先前对运动估计的研究主要集中于快速的整像素运动估计算法,对于快速模式选择算法的研究很少。本文中,我们分析了运动估计不同阶段下最优模式之间的相关性,并在此基础上提出了快速模式选择算法。该算法可简化模式选择的过程,并在很大程度上降低亚像素运动估计的运算量,所带来的编码损失基本可忽略不计。在算法级对H.264进行了优化后,我们又利用Intel的SSE-2多媒体指令集对H.264编码器中的运算密集模块进行了进一步优化,大幅度提高了编码器的速度。最后,我们在前面工作的基础上,开发了基于上下文切换技术的多线程H.264编解码函数库,支持多通道同时编解码。在使用Intel Pentium 4 CPU、256M SDRAM的PC平台上,单通道的编码速度在QCIF分辨率下可达40帧/秒以上,在CIF分辨率下可达10帧/秒。更多还原

【Abstract】 H.264 is a new video coding standard jointly developed by ITU-T and ISO MPEG. It has drawn a lot of attention because of its excellent coding performance. However, the high computational complexity of H.264 holds it back from practical uses. In this paper, we’ll focus on optimizations of H.264 encoder at algorithm level and instruction level. Motion estimation is the most important module in a video encoder. The algorithm can directly impact the speed and performance of the encoder. H.264 introduces 7 block size modes, and improves the motion compensation accuracy to 1/4-pixle. Previous studies on motion estimation mainly focus on integer-pixel motion estimation, and very few deal with fast mode decision. In this paper, we’ll analyze the correlation between the best modes of different accurate levels(integer-pixel、1/2-pixel、and 1/4-pixel), and based on this, we propose a new fast mode decision algorithm. This algorithm can simplify the procedure of mode decision and significantly reduce the computational complexity of sub-pixel motion estimation. The loss of coding performance is negligible.After optimizing H.264 encoder at algorithm level, we’ll further optimize the computationally intensive modules in H.264 encoder using Intel’s SSE-2 instruction set and significantly speed up the encoder.Finally, based on works in previous chapters, we develop a multi-thread H.264 Codec static library. The Codec uses context-switching technique and supports simultaneous encoding of multiple channels. In a PC with Intel Pentium 4 CPU and 256M SDRAM, the single-channel encoding speed is above 40f/s for QCIF sequences and 10f/s for CIF sequences.更多还原