【作者】 周冬辉；

【导师】 端木春江；

【作者基本信息】 浙江师范大学 ， 计算机软件与理论， 2010， 硕士

【摘要】 最新的国际视频编码标准H.264是由ITU-T视频编码专家组(VCEG)和ISO/IEC动态图像专家组(MPEG)组成的联合视频组(Joint Video Team, JVT)在2003年提出的。因为H.264采用了许多新的技术,如：可变块大小的运动估计(variable block size motion estimation)、小数像素的运动估计(decimal pixel resolution motion estimation)、多参考帧的运动估计(multi-reference frame motion estimation)、率失真优化(rate distortion optimization, RDO)、CABAC(context-based adaptive binary arithmetic coding)、CAVLC(context-adaptive variable length coding)等,H.264的编码效率要优于以前的视频编码标准。然而也因为采用了这些技术,使得H.264的运动估计部分的计算复杂度大大高于以前的编码标准。正因为如此,降低H.264中运动估计复杂度的算法的研究是近年来国际上的研究热点。在充分考虑了各种图像序列的运动特性的基础上,我们提出了基于矢量预测和多方向梯度下降搜索算法(Multi-Direction Gradient Descent Search,MDGDS)算法。该算法首先利用运动矢量的时间和空间上的相关性来得到预测矢量的搜索起始点,再通过使用自适应阈值来判断当前块的运动类型,以此来采用不同的搜索策略。这个算法可以快速扩展搜索范围和提高运动估计的搜索速度,从而避免使搜索过程陷入局部极小。实验结果表明,与非对称十字多级六边形搜索算法(UMHexagonS)、简化的非对称十字多级六边形搜索算法(简化UMHS)以及增强预测区域搜索法(EPZS)等传统算法相比,本算法能在保持运动估计精度的同时进一步地节约大量的编码时间。在通过对多个QCIF视频序列的编码结果的统计分析的基础上,我们发现以下三个特征：每种模式被采用的概率不均匀；相邻宏块的模式与当前宏块的模式存在着比较大的相关性；上层模式和下层模式之间存在一定的相关性。同时,由于模式的分布并不均匀,在某些情况下,我们可以忽略一些对编码效率的影响很有限和出现几率很小的模式。基于上述发现,我们提出了一种基于统计特征的适于H.264的快速模式选择算法。该算法利用编码模式的上述统计特征来提前终止运动估计中模式的搜索过程,从而减少运算量和提高编码效率。实验结果表明,与高复杂模式(High complexity mode, HCM)算法、以及快速高复杂模式选择(Fast high complexity mode, FHCM)算法相比,所提出的算法能在保持几乎相同的率失真性能的同时进一步地节约大量的编码时间。本文的结构如下：首先介绍了H.264视频编码的原理,其中着重描述了H.264中的运动估计的原理和关键技术。接下来,详细描述了JM15.1(H.264的参考软件包)中的整数像素快速运动估计算法和快速模式选择算法。然后,重点阐述了我们所提出的一种新的整数像素快速运动估计算法和一种新的快速模式选择算法。更多还原

【Abstract】 New video coding standard H.264 was published by ITU-T Video Coding Experts Group (VCEG) and the ISO/IEC Moving Picture Experts Group (MPEG) formed a Joint Video Team (JVT) in 2003. Because some advanced coding technologies are used in H.264, such as variable block size motion estimation, fractional pixel motion estimation, multi-reference frame motion estimation, rate-distortion optimization (RDO), context-based adaptive binary arithmetic coding (CABAC), context-adaptive variable length coding (CAVLC) etc, H.264 gain more coding efficiency than other video sequence standard. However these technologies are employed, the motion estimation of H.264 is much higher computation complexity than previous video coding standards. Because of these reasons, the study of motion estimation algorithm is always hot spot research in recent years to reduce the computational complexity of motion estimation in H.264.Based on the consideration of various motion characteristics of different video series, we propose a motion estimation algorithm based on vector prediction and multi-direction gradient descent search. Firstly, the algorithm utilizes the temporal and spatial correlations of the motion vectors for obtaining a predictive vector to determinate initial search point. Secondly, it determines the motion type of the current block by using adaptive thresholds. After that, different search patterns and strategies are intelligently employed. The algorithm can extend the search scope and improve the speed of motion search, and thus avoid being trapped in a local minimum of the search process. Experiment results show that the proposed algorithm can save more encoding time than the traditional hybrid Unsymmetrical-cross Multi-Hexagon-grid Search (UMHexagonS), simplified UMHexagonS and Enhanced Predictive Zonal Search (EPZS) algorithms with the motion estimation accuracy.Based on the statistical analysis of extensive experiment results of various QCIF video test sequence, we observed three statistical features, such as the distribution of modes is not uniform; The mode of a macroblock and the mode of macroblocks neighboring have strong correlation; the uplayer mode and dwonlayer mode have a certain correlation; At the same, because the distribution of modes is not uniform, in some cases, we can ignore some modes that have very little effect on encoding efficiency and are almost not used. Based on the analysis above, we propose a fast mode decision algorithm in H.264 base on statistical feature. The algorithm takes advantage of the above statistical features to early termination of mode decision, so it can reduce computation load and improve efficiency of video coding. Experiment results show that the proposed algorithm can save more encoding time than the mode decision algorithm of High Complexity Mode (HCM) and Fast High Complexity Mode (FHCM) with negligible degradation of rate-distortion performance.The paper is organized as follows. Firstly, principle of video coding in H.264 are briefly introduced, principles and key technologies of motion estimation in H.264 are detailed introduced. Secondly, Fast integer pixel motion estimation algorithms and fast mode decision algorithm in JM15.1 (H.264 Reference Software) are detailed. Lastly, a new fast integer pixel motion estimation algorithm and a new fast mode decision algorithm are proposed.更多还原