资源预览内容
第1页 / 共56页
第2页 / 共56页
第3页 / 共56页
第4页 / 共56页
第5页 / 共56页
第6页 / 共56页
第7页 / 共56页
第8页 / 共56页
第9页 / 共56页
第10页 / 共56页
亲,该文档总共56页,到这儿已超出免费预览范围,如果喜欢就下载吧!
资源描述
I摘摘 要要 运动估计是视频信息处理中的关键技术, 主要目的是获取视频图像的运动信息,便于视频信息的应用。由于运动估计的计算量通常在视频处理系统中占总计算量的60-80%,运动估计算法直接影响到系统的效率,因此寻找一种简便、快速、高效的运动估计算法成为视频信息处理领域的一个研究热点。在过去的研究中,研究者虽然提出了许多运动估计算法,以提高估计性能。但是,由于块匹配运动估计实现简单,因此对其研究受到特别关注,出现了以运动矢量均匀分布为基础的矩形模型块匹配运动估计算法, 以及以运动矢量分布中心偏置特征为基础的非矩形模型估计算法。其中后者能较大程度地提高搜索速度,但由于搜索模型的复杂性,实现比较困难,特别是对于硬件实现。此外,由于视频对象通常为任意形状,不能满足搜索模型的要求。针对这些问题,以运动矢量概率分布特点为基础进行运动估计算法研究。 运动矢量概率分布特性分析是获得最佳运动估计性能方法之一, 因此使用全搜索算法对包含不同运动内容视频序列的运动矢量分布进行测试, 发现了现实视频序列的运动矢量分布除了具有中心偏置特性之外,还具有方向性偏置分布特性。 该特性反映了现实运动序列的运动矢量在搜索窗口的中心坐标轴方向上的概率分布比对角线方向上的概率分布大。 根据运动矢量概率分布的中心偏置性和方向性偏置性分布, 提出了针对不同运动矢量的搜索模型。不同形状、大小的搜索模型很大程度上决定了运动估计搜索的效果和速度,根据视频序列运动矢量的偏置特性,给出了对于中心偏置特性的大/小十字形搜索模型(L/SCSP),以及对应于方向性偏置特性的新的非对称十字形搜索模型。该模型相对于以前的菱形、十字等搜索模型更符合现实视频序列中的运动规律。 提出一种新十字形块匹配运动估计算法。该算法在搜索策略上对大、小运动矢量的搜索区别对待。在搜索初始阶段同时使用小十字搜索模型(SCSP)和大十字搜索模型(LCSP)对小运动矢量进行搜索,尽可能通过减少搜索所需要的点,加快对静止或准静止块的搜索。在后继的搜索步骤中,充分考虑大运动矢量概率分布的方向性,采用非对称十字搜索模型进行搜索,从而加快对大运动矢量的搜索速度。该算法保持与菱形搜索(DS)和十字-菱形搜索(CDS)算法相当的搜索质量下,可提高大约60%和35%的搜索速度。 根据MPEG-4校验模型(VM:Verification Model) ,结合新十字形块匹配运动估计算法,实现了基于对象的运动估计算法的编码器。该编码器将视频对象的宏块被分为3类:对象内部宏块,对象边缘宏块和对象外部宏块。 对于对象内部宏块,采用新十字形块匹配法进行运动估计,对于对象外部的宏块则不进行运动估计,而对象边缘的宏块(部分像素在对象内部)就采用填充的方法,利用多边形匹配技术进行运动估计。 关键词:运动估计;视频编码;视频对象;运动矢量偏置特性 IIIAbstract Due to the fast development of communication technique, transmission of a video sequence has been emphasized in the near future. Considering the limited channel bandwidth and real-time process requirement, it is necessary to apply efficient video source and high compression ratio coding method. For motion estimation plays an important role in inter-frame predictive coding system to realize the high video compression ratio, it draws a lot attention of researchers. Recently, researchers have proposed many kinds of motion estimation algorithms in which the block-matching motion estimation is paid particular attention for its simple implementation. Therein the rectangular-block matching motion estimation algorithm which is based on the factor of motion vector probability distribution. and the non-rectangular block estimation algorithm which is based on the factor of motion vector center-biased distribution are two kinds. The later can improve the searching speed, but due to the complexity of searching model, it is difficult to implement, especially in hardware. Furthermore, video object can not very well fit the requirement of searching model because of its typically arbitrary shape. Due to these problems, the research of motion estimation algorithm is mainly based on the factor of motion vector probability distribution. One way to gain the best performance of motion estimation is to analyze the factors of motion vector probability distribution. Therefore, the Full-Search algorithm was used to test the distribution of motion vectors including video sequence of different motion content. During the research, we found that the actual video sequence distribution had not only the centre-biased factor but also the direction-biased factor. The two factors expressed that in the actual motion sequence the probability distribution of motion vectors in the direction of centre coordinate-axis was larger than the direction of diagonal. According to the center biased and direction biased of the motion vectors distribution, we proposed many different search patterns. In block motion estimation, search patterns with different shapes or sizes have a large impact on the search speed and the performance quality. According to the bias of the motion vectors distribution, the Large cross search pattern and Small cross search pattern based on center-biased property, and a new search cross pattern based on direction-biased property are proposed. These patterns can reflect the real motion vector distribution better in most video. In the paper, we propose a new cross algorithm. It differs from other BMAs. It is suggested to firstly employ a Large/Small cross-shaped pattern to explore cross-center-biased property of most real word sequences, then detects search direction with an unsymmetrical cross search pattern in subsequent step. Simulation shows that NCS can improve the search speed by about 60% and 35% compared with the Diamond Search (DS) and Cross Diamond Search(CDS) algorithm. According to the MPEG-4 Verification Model, an object-based motion estimation algorithm which is based on our new cross search is implemented in a codec. This codec divide the video object macro-blocks into three types: macro-block inside the object, macro-block on the edge of the object and macro-block outside the object. For the inside macro-block, a new cross search estimation algorithm is used. For the outside macro-block, no motion estimation is used, and for the on-edged macro-block, the padding algorithm is adopted with polygon technology. Key Words: video encode, motion estimation, video object, motion vector probabilities distribution characteristics . - - 1第 1 章 绪 论 1.1 引言 第 1 章 绪 论 1.1 引言 当今世界已跨入信息时代,信息的形式也从单纯的语音发展至包括数据、文字,视频图像的多媒体信息。其中,视频信息以直观性强,信息量大及便于理解等特点占有举足轻重的
网站客服QQ:2055934822
金锄头文库版权所有
经营许可证:蜀ICP备13022795号 | 川公网安备 51140202000112号