【作者】 周城；

【导师】 田金文； 谭毅华；

【作者基本信息】 华中科技大学 ， 模式识别与智能系统， 2007， 硕士

【摘要】 随着近年来多媒体技术的飞速发展,许多应用领域对视频图像的实时压缩提出了更高的要求,快速、高效的压缩算法是解决这一问题的关键。H.264是ITU-T VCEG组织和ISO/IEC MPEG组织共同制定的新一代视频编码标准。虽然H.264采用了传统的混合编码框架,但通过使用包括帧内预测、整数变换、可变预测块大小、多参考帧、1/4像素精度预测、自适应算术编码、去块效应滤波以及基于率失真优化的模式选择等新技术,编码效率获得了很大提高。AVS(Audio Video Coding Standard)是目前我国正在制定的具有自主知识产权的音视频编码技术标准,它的编码效率与H.264相当,其优势为:性能高、复杂度低、实现成本低、专利授权费用低。运动估计是视频编码中关键技术之一。本文对H.264和AVS中的运动估计技术进行了深入分析,并提出了相应的优化算法。UMHexagonS整像素快速运动估计算法性能优秀,先后被JVT和AVS接受为标准算法。本文在对UMHexagonS算法进行了详细分析后,提出了优化算法。同原算法比较,能够在性能仅下降0.03dB的情况下,节约30%~90%的运动估计时间。AVS采用了可变大小块以及1/4像素精度的运动估计技术以提高编码效率。本文基于AVS参考软件提出了一种增强亚像素自适应范围搜索算法。该算法主要使用了小菱形窗的搜索策略和阈值判决方法。与亚像素全搜索算法比较,在平均性能仅下降0.0093dB的情况下,减少了30.25%的搜索点数。随着数字音视频产业的发展,嵌入式视频压缩技术应用越来越广泛。本文阐述了基于TMS32DM642 DSP平台的通用视频编码器优化方法,包括系统算法优化、存储器的分配、使用EDMA传输数据和调用TI库函数等。同时阐述了对基于MPEG-4平台的PMVFAST快速运动估计算法的优化方法。基于这些方法,在DM642平台上采用MPEG-4编码器对CIF格式图象进行编码,帧率可达40fps,完全满足实时编码的要求。更多还原

【Abstract】 With the fast develop of multimedia techniques, there are many areas need more efficient real-time compression techniques for digital image processing. Fast compression algorithm is the key for this problem. H.264 is the current project of the ITU-T Video Coding Experts Group (VCEG) and the ISO/IEC Motion Picture Experts Group (MPEG). Although H.264 follows the hybrid coding scheme in common with earlier standards, improvements are achieved through the inclusion of a number of new features that distinguish itself from all existing ones, including Intra Prediction, Integer transform, different predicted block, muti-conference frames, CAVLC,CABAC, quarter-pel precision prediction, Deblocking filter, R-D mode strategy, etc. The being finished Chinese audio and video coding standards AVS has approximate coding efficiency as H.264, AVS’s advantages includes high performance, low complexity , low implementation price and sample licensing.Motion Estimation is a key technique for video coding. This thesis is concerned with the Motion estimation techniques of H.264&AVS and proposes improved methods in several fields. UMHexagonS for Fast Integer Motion Estimation is adopted in previous JVT and AVS meetings because it achieves a very good trade-off between ME acceleration and R-D performance. This thesis present some characters within UMHexagonS process and makes corresponding improvements. Compared with the original technique, 30~90% ME time is reduced while the limited performance lose about 0.03dB. AVS applied the variable block sizes and quarter-pel motion compensation to improve coding efficiency. In this paper, an adaptive range fractional pel search algorithm is proposed to further speed up the encoding process and reduce the computation complexity. Small diamond search and threshold judgment are used to this algorithm. Compared to the half-pel full search algorithm, experiments show that the proposed algorithm can reduce the sub-pel search point by 30.25% on average with the limited performance lose about 0.0093dB.With the developonent of the digital audio and video industry, embedded video compression technology is applied more widely. This thesis describes the optimization techniques for a general video encoder based on TMS32DM642 DSP. The algorithm optimization, memory distribution, EDMA and TI’s function libraries usage are discussed based on DM642. And the optimization technique for the PMVFAST Fast Integer Motion Estimation base on MPEG-4 are provided. An MPEG-4 encoder based on DM642 using these methods which encodes images of CIF can carry out real-time encoding with 40fps.更多还原