【作者】 于新波；

【导师】 赵辉；

【作者基本信息】 山东大学 ， 信号与信息处理， 2008， 硕士

【摘要】 随着现代数字通信技术和信息处理技术的不断发展,以及数字电视、网络视频流和DVD技术的广泛应用,视频压缩技术越来越受到人们的重视。视频压缩作为计算机多媒体技术的一个重要分支,近年来正逐渐成为国内外研究的热点。视频压缩的最根本目的是以尽量少的比特数来表征视频流,同时保证视频的主观质量,使其符合各种应用场合的要求。因此,探讨高效的编码方法,即在提高压缩比的同时以更低的计算复杂度来获得更高质量的图像,一直是视频编码研究的热点问题。为此,本文在研究多媒体技术发展的现状以及目前各种视频压缩技术的基础上,分别对视频压缩中的预处理和运动估计算法进行了详细研究,提出了相应的改进方案,并经实验验证了其有效性。主要工作和研究成果包括:1.对视频压缩中的预处理方式进行了研究。人眼是视频信号的最终接收者,因此视频的压缩编码应当考虑人眼的主观视觉特性。由于人的视觉能力有限,对有些失真很难察觉,因此只要失真在人眼视觉系统可容忍的范围之内或者能够满足特定的应用要求,便可以允许压缩编码后的重建图像存在一定程度的失真。基于此观点,本文通过利用人眼的视觉特性,对视频图像进行适当优化处理,将视频压缩时会产生较大失真的敏感部分进行抑制,再进行正常的压缩编码,从而在一定程度上实现图像压缩比的提高和图像主观质量的改善。实验结果表明,本文提出的预处理算法在提高压缩比的同时可以有效保证视频的主观质量;在对原始视频进行较大程度压缩时,采用预处理后的解码图像主观质量明显好于未经处理而压缩的结果。2.对视频压缩中的运动估计算法进行了研究。运动估计作为视频压缩的核心技术,是压缩编码中运算最复杂,也是最耗时的环节,因此高效的运动估计算法一直是实时视频编解码技术研究的重点。为降低运动估计算法的计算复杂度,提高运动估计效率,本文首先对传统的快速搜索算法以及ARPS算法进行了研究,然后从四方面对APRS算法进行了改进:利用邻块的运动矢量来预测当前块的搜索起点;引入基于运动矢量中心分布的初始搜索;采用自适应的非对称的十字形搜索模板;结合早期终止机制提前结束搜索。实验结果表明,与TSS、DS和ARPS等搜索算法相比,改进后的算法不仅能够降低计算复杂度,而且可以有效地保证搜索精度。另外,本文对H.264标准JM中采纳的快速运动估计算法UMHexagonS也进行了相应的研究,针对其中的非对称十字形模板进行了改进,提出了一种新的自适应的十字形搜索模板,并在JM10.2中进行了实验验证。实验结果显示,采用改进模板的搜索算法不仅可以提高运动估计效率,而且可以有效保证甚至改善编码图像的质量。更多还原

【Abstract】 With the rapid development of modern digital communication technology and the extensive applications of HDTV, Net Video and DVD, video coding techniques are highly regarded now. As an important branch of computer multimedia technologies, video coding is becoming a focus of academic research all over the world.The goal of digital video compression is using fewer bits to represent video stream.And at the same time it should keep the subjective quality of the compressed video good, and make it satisfy specific applications. Thus, to find more efficient schemes which can improve compression rate and visual quality with lower coding complexities is always the goal of video coding research.Based on the research of the development and basic techniques of video coding, the thesis focuses on video pre-processing methods and motion estimation algorithms.The main achievements of this thesis include:1. Pre-processing methods for video compression.For the video signal, human eyes are the final receivers, thus video compression should consider the effect of human visual system. As the ability of human eyes is limited, it is hard to detective some distortions.Thus, if the distortion can be tolerated by human eyes or it can satisfy some applications, video codec can allow certain distortions to appear in the reconstructed images. Before compression and transmission, we can take advantage of the characters of HVS to pre-process original video sequences, and then encode and transmit them. Experimental results show that the proposed algorithm can improve compression ratio to a great extend without sacrificing subjective video subjective quality, and at the same compression rate, the subjective video quality using pre-processing technique is obviously better than that of the original compression scheme.2. Fast search algorithms for block-matching motion estimation.As a key part of video coder, motion estimation is extremely complicated and time-consuming.Thus an efficient motion estimation algorithm is always essential to the research of real-time video coding and decoding.Based on adaptive rood pattern search (ARPS) algorithm, a new fast and efficient motion estimation algorithm called enhanced adaptive rood pattern search (E-ARPS) is proposed in this thesis. Four aspects are proposed to improve ARPS: utilize spatial correlations to predict the initial search centre, introduce a new initial search based on motion vector (MV) centre-biased distribution, adopt a new adaptive unequal-arm rood search pattern and incorporate early terminations. Experimental results show that our algorithm has yielded significant search efficiency gain while consistently ensure search accuracy. Based on the study of the principle of the UMHexagonS algorithm, a new adaptive unsymmetrical-cross pattern is adopted instead of the unsymmetrical-cross proposed in the UMHexagonS. In JM10.2, experimental results show that the improved UMHexagonS algorithm can effectively reduce motion estimation duration and improvethe quality of the reconstructed images compared with the original performance of the UMHexagonS algorithm.更多还原