【作者】 张博；

【导师】 戴琼海；

【作者基本信息】 清华大学 ， 控制科学与工程， 2005， 硕士

【摘要】 随着无线网络的迅猛发展,无线视频媒体成为一个应用研究热点。最新的H.264视频编码标准兼具良好的压缩效率和网络适应性,非常适合成为无线媒体应用中的标准格式。然而,具有高压缩率的同时,它也带有极高的计算复杂度。采用ITU提供的参考软件,在无线环境的服务器端和客户端都远不能达到实时应用,因此有必要针对无线应用环境进行复杂度优化,提高编解码速度。复杂度优化可以分为算法优化、与机器无关的代码优化和针对体系结构的优化。由于无线视频的编码端异构性较强,编码流程有可能随应用变化而改动,并且编码算法复杂,有算法优化的余地,因此对编码部分采取算法的优化。在考查编码中各个环节的复杂度,分析了编码流程瓶颈之后,将优化的重点放在了帧间预测模式快速选择算法上。通过对当前已有的多种快速预测算法进行研究,将其分为简化RD优化过程的方法、通过运动补偿结果判断的快速模式选择、基于空间信息的快速算法和基于简单时间信息的模式选择算法。在研究的基础上,提出了一种合理的基于时间信息的优化算法,通过P帧参考图像的分块结果,考虑子块的运动特征,对当前图像进行快速帧间模式选择。由于在算法中增添了匀速直线运动的假设,所以算法同时包含场景变换的快速判断,以消除假设条件不成立所带来的压缩效率损害。经过仿真验证,此算法可以使运动搜索环节的计算时间节省40%,同时PSNR下降不超过0.1dB,码率增加不超过2.5%。在解码器中,由于算法的灵活性较差,可优化幅度有限,主要采用了针对体系结构的优化方法。通过对无线应用的环境进行分析,选择XScale系统作为典型应用平台。通过研究XScale的MAC、内存结构、缓存特性和流水线等特点,并通过对解码器的瓶颈进行分析,确定出优化策略。通过改变内存访问方式、微调子程序结构、锁住数据缓存和进行汇编优化,使解码性能大幅度提升,达到实时解码播放的程度。更多还原

【Abstract】 With the rapid development of wireless networking and mobile systems,video streaming in wireless environment has attracted more and more focusand research. H.264, the newest video standard, is suitable for wirelessenvironment, because it has very high compressing quality and can be robustin network transmission. However, it has a rather high computationalcomplexity in the encoding and decoding process. It is difficult to performrealtime application with the reference software provided by ITU. So we haveto optimize the process in the encoder and decoder to decrease the complexity. By analyzing the encoding process, we can find that the inter prediction isthe most time consuming module, and the multi inter modes adopted by H.264intensify the complexity greatly in inter prediction. Several fast modedecision methods have been brought forward, most of which are concernedwith the spatially information in the current picture, and some are designatedto simplify the RD process. After researching and analyzing the methods, anew scheme on temporal information is proposed. The results of interprediction in the reference P pictures are taken into consideration, not only thevariable block size modes, but also sub blocks’ moving tendencies. With theirmoving status, the small blocks are reconstructed in the frames to be coded,and then the mode decisions can be made. Experiment results have shown thatthe new method will decrease the times of motion estimation at about 40%with negligible impact on coding performance. As for the decoder in wireless video environment, it is difficult todecrease the complexity by changing the decoding methods, because they arealmost fixed. So the work mainly forcus on adapting the decoding process to aspecial system architecture. Decoders with XScale CPU and WinCE OS areselected as the most typical application environment. The memoryarchitechure, cache features, and pipeline abilities are examined, and a seriesof optimization schemes are brought forward. By rearranging the memory,changing the flow of some decoding modules, locking data cache and usingassembly language to perform the function in certain modules, the decodingspeed increases remarkably and realtime decoding has been achieved.更多还原