【作者】 杨铭；

【导师】 崔慧娟；

【作者基本信息】 清华大学 ， 信息与通信工程， 2004， 硕士

【摘要】 随着软硬件处理能力不断提高,人们对多媒体信息需求日益增加,导致近二十年中数字视频技术得到了飞速的发展。不同厂商提供的视频码流的互通互联需求促使了国际视频压缩标准的出现,反过来极大地推动了许多数字视频应用领域的发展,如数字电视、互联网流媒体、数字存储个人娱乐影片、视频会议和可视电话等。H.264/AVC是ITU和ISO/IEC联合制订的最新国际视频压缩标准,其中众多新颖的算法特性大幅提高了编码性能,同时运算复杂度也较其他视频标准有巨大增长。如何尽量在保持H.264编码性能的基础上,减少其运算复杂度提高运行速度,是H.264能否实际应用的关键问题。本论文针对软件实现H.264编码器问题,在分析H.264各部分运算复杂度和对编码性能贡献的基础上,针对H.264中树状结构运动估计的特点,首先提出了连续消除菱形运动矢量搜索快速算法,使整象素搜索中块匹配运算量减少至全搜索0.2%~1%,代价为图像平均PSNR下降0.1dB平均码率增加1%以内;进一步提出基于宏块运动内容分类的快速宏块编码模式选择算法,减少了40%-70%的H.264块匹配计算量;进而研究了Pentium-4平台上,采用MMX/SSE2汇编语言进行SIMD优化和其他优化措施,基本实现针对QCIF和CIF图像格式序列的H.264软件编码器。算术编码传输中对误码高度敏感,本论文对H.264中基于上下文的自适应二进制算术编码的实现方法进行了研究,针对有限精度的算术编码器,提出了一种添加相同程度冗余信息,使用多个禁止符号的算术编码检错算法,具有较强的检错能力,错误符号立即检错比例提高了1倍,对算术编码联合译码有很大帮助。论文的成果已经部分应用于实际无线视频通信系统中。论文提出的快速和软件优化方法对多媒体软件算法优化有一定参考价值,对H.264视频标准的应用具有重要意义。更多还原

【Abstract】 With the rapid progress of hardware and software processing ability and increasing requirement of multimedia information, digital video technology has achieved great development over the last two decades. The demands for the interconnecting capability and interoperability with video streams supplied by different vendors boost the emergence of international video compression standard. In the meantime, the standardization greatly promotes many new application fields of digital video, such as digital television, Internet multimedia streaming service, interactive digital storage for personal entertainment and video conference.H.264/AVC is the newest international video coding standard, joint developed by ITU and ISO/IEC, a great many of novel algorithm components enhance the coding efficiency significantly, while dramatically increase the computational complexity than the previous standards. How to reduce its complexity and accelerate the execution speed while maintaining the outstanding compression performance is a crucial problem in the H.264 application. This dissertation mainly addresses the software implementation problem of H.264 encoder. Based on the analysis of various algorithm compositions’ computational complexity and the contribution to the compression efficiency, a novel fast motion estimation algorithm focusing on the tree structured motion estimation in H.264, successive elimination diamond search (SEDS), and a new fast macroblock mode selection (FMMS) method based on the classification of macroblock motion content are proposed. The proposed fast algorithms reduce the block matching calculation substantially while maintaining similar coding efficiency. Simulation results demonstrate that the block matching calculations in full-pel motion estimation are reduced to 0.2%-1% by employing the SEDS algorithm, and further reduced by 40%-70% with FMMS. Furthermore, the dissertation studied the important SIMD <WP=5>optimization methods with MMX/SSE2 assembly languages and other optimization means on Pentium-4 platform. A practical H.264 software encoder is implemented which can encode QCIF and CIF format image sequences in real-time.The bitstreams of arithmetic codes are very vulnerable to transimission errors, so the implementation methods of one of the entropy coding options in H.264 standard, the context-based adaptive binary arithmetic coding, are studied in the last chapter of the thesis. With the analysis of the finite precision implementation of the arithmetic coder, a new error detection method using multiple forbidden symbols is proposed. With the same amount of redundancy, the new method can achieve better error detection capability than the previous single forbidden symbol method in terms of the bit and symbol detection distances, which is very helpful for joint decoding of arithmetic codes.Parts of the achievements have been applied to some practical video communication systems. The fast algorithms and optimization methods in this dissertation can be referenced by other multimedia software algorithm optimization, and have practical significance to the application of H.264/AVC video coding standard.更多还原