【作者】 陈国斌；

【导师】 刘济林；

【作者基本信息】 浙江大学 ， 通信与信息系统， 2003， 博士

【摘要】 数字视频是多媒体的重要标志，也是多媒体信号中最具有表现力的成分。但数字视频内在的高带宽特性限制了多媒体业务的发展。为有效利用传输带宽和存储空间，必须对视频数据进行压缩。ISO／IEC和ITU-T分别制定了MPEG-x和H.26x系列标准以便于对数字视频的处理、传输和存储。除最新的H.264外，这些标准都是基于联合MC-DPCM／DCT的视频压缩框架，即采用运动补偿预测和离散余弦变换分别消除视频信号的时域和空域冗余。 视频编码是在码率、质量、处理复杂度、容错性能和交互性等约束条件下的多维优化过程。本文主要从实时视频应用角度，对混合MC-DPCM／DCT视频编码中的处理复杂度、码率控制和差错控制这三方面问题展开了深入的研究。 众所周知，视频处理是多媒体计算中运算开销最大的部分，它的计算复杂度远大于其它媒体(文本、语音、音频、图形和静态图象)。对于实时视频应用环境，视频编码以往大多由专用设备完成。但随着通用处理器和数字信号处理器(DSP)主频的提高、面向视频处理的指令集出现，使得更为经济、灵活的软件编码成为可能。本文系统分析了MC-DPCM／DCT视频编码原理，从算法和实现这两方面探讨了降低视频编码复杂度的方法，首先重组了编码器结构；接着对编码过程中的关键模块进行优化。特别地，针对移动视频通信应用，本文提出了一种低复杂度的高性能运动估值算法；然后将与处理器硬件结构相关的方法引入了优化过程，提高了数据的处理并行性和访问效率；最后使用查找表或汇编优化对关键模块进行加速。 其次，在码率控制方面，码率—失真(R-D)分析和码率控制在视频编码和视频通信系统中起着至关重要的作用。在传统R-D模型中，码率R和失真D都看成量化步长的函数，而且失真评价函数一般在象素域或空域中进行。本文在压缩域或DCT域建立了一种失真函数模型，它能够对编码复杂度提供更加精确和可靠的预测。在此基础上，针对TM5模型中存在的缺陷，本文提出了一种改进的CBR码率控制算法，它解决了TM5中缓冲区不一致性问题，并增强了算法对场景切换的适应性。对于VBR码率控制，作者还提出了一种面向实时视频编码应用的单通道VBR码率控制算法，它避免了多通道VBR算法的多次编码或预分析的开销，减小了量化步长波动，可提供稳定的视频质量。 最后，在视频通信系统中，由于传输信道中噪声或网络中拥塞影响，传输差错和数据包丢失是不可避免的。与普通的数据传输不同，视频通信受严格的延时限制，不能通过重传机制保证无差错的传输。预测和可变长编码的使用也使得码流对误码或数据包丢失更加敏感。本文在分析MC-DPCM／DCT压缩视频特点的基础上研究了各种误码检测方法，并详细讨论了编码端的误码容错、传输层差错控制、解码端的失真屏蔽以及编码器和解码器交互的差错控制方法。更多还原

【Abstract】 Multimedia is characterized by its video capability since video is the most efficient presentation method among various digital media signals. But digital video has an inherently high bandwidth, which limits the video application in multimedia services. To maximize the use of channel and storage, video signal need to be efficiently compressed. The increasing demand of video interoperability ignited ITU-T and ISO/IEC to develop lots of standards for video processing, transmitting and storing. Except H.264, almost all of the standards use MC-DPCM/DCT technologies. Motion compensation and DCT (discrete cosine transform) are used to decorrelate spatio-temporal correlation of video signal and quantization is used to exploit perceptual redundancy of nature video.Video coding can essentially be view as a multi-dimensional optimization problem; the key dimensions concerned in this problem are rate, quality, coding computational complexity, error resilience and memory size. In order to provide insights into common video system design problems, several issues for real-time video coding have been addressed in this dissertation, which are video coding computational complexity reduction, rate control, error control and error resilience.It is well known that video processing is the most computationally intensive part in multimedia computation. Real-time video coding previously resorted to dedicated hardware or professional equipments. With the development of general-purpose processors and high efficient algorithms, it is possible to implement a software-based real-time video encoder, and its low cost and easy upgradability attract developers’ interests to migrate video encoding from dedicated hardware to more flexible software. In this paper, we first rearrange the encoder structure to support complexity scalability; then a lot of high performance algorithms are used on the key time-consuming modules in coding process; next, in programming level, processors characters are considered to improve data access efficiency and processing parallelism; finally, other programming methods, such as lookup table, explicitly assemble on key modules, are used to speed up coding process.Secondly, rate-distortion (R-D) analysis and rate control play an important role in video coding and communication systems, which are to prevent buffer malfunction and provide the highest possible video quality under the constraints of rate and delay. In conventional R-D analysis, the bitrate R and distortion D are considered as function of quantization parameter q, which are usually developed in pixel-domain or spatial domain. These source models either have high computational complexity or subject to relatively large estimation evaluation and control error. In this dissertation, a R-D analysis in DCT domain or compressed domain is developed, which can provide more accurate prediction for target bit allocation. Based on DCT domain R-D analysis, we provide a new constant bitrate (CBR) rate control algorithm, which can cure TM5 algorithm’ two limitations, buffer conformance and scene cut problem. For variable bitrate (VBR) rate control, weproposed a single-pass VBR rate control algorithm for real-time video coding application. It doesn’t require pre-analysis and multi-pass coding, and can achieve a constant video quality.Thirdly, in video communication system, transmission errors and packet losses are inevitable due to channels noise and network congestion. In contrast with data communications, which are not usually subject to strict delay constraints and can therefore be handled using network protocols that uses retransmission to ensure error-free delivery, real-time video is delay sensitive and cannot easily make use of retransmission. The extensive use of prediction and variable-length coding in video coding renders compressed bitstreams especially vulnerable to transmission errors and data losses. Based on detailed analyses of MC-DPCM/DCT mechanism, several error detection, resilience, concealment methods are discussed in this paper.更多还原