【作者】 纪旭；

【导师】 苏育挺；

【作者基本信息】 天津大学 ， 信号与信息处理， 2008， 硕士

【摘要】 伴随着多媒体技术和网络技术的快速发展,大量的视频信息不断涌现,使得对视频数据的有效管理、浏览和检索成为了亟待解决的问题。与一般的视频数据相比,新闻视频有着特殊的四层结构:视频帧-镜头-故事单元-整个视频。新闻视频的静态摘要技术也因其在视频的浏览、检索和传输方面的应用而受到了广泛地研究。论文首先描述了新闻视频的结构以及视频处理中常用的视觉、音频、文本特征以及压缩域特征。然后简要介绍了镜头边界检测、关键帧提取和故事单元分割三种结构分析技术,其中的关键帧提取技术是静态视频摘要的核心。接着,详细介绍了本文提出的三种新的关键帧提取技术,分别为:基于自适应阈值聚类的方法、基于协方差的方法以及基于条件熵的方法。基于自适应阈值聚类的方法利用图像分割技术中的分水岭算法和Otsu算法来计算自适应的阈值。基于协方差的方法以及基于条件熵的方法则是利用了相邻视频之间存在着高度相关性的特点,尽量减少提取的关键帧之间的冗余。实验结果证明了三种方法的有效性。为了满足不同人群的观看需要,还提出了一种基于分层的静态视频摘要方法,该方法根据人的主观感受来决定提取关键帧的数量。最后实现了一个基于COM架构的静态视频摘要系统。更多还原

【Abstract】 With the rapid development of multimedia and network techniques, effective video management, browsing and retrieval become more and more important due to the large amount of information video provided. News video has its own structure: frame-shot-story-video. The technique of news video summary has been widely investigated because of its applications in video browsing, retrieval and transmission.In this dissertation, the structure of news video is described and some video analysis features, such as visual feature, audio feature, text feature and MPEG domain feature in video processing are also introduced first. Second, shot boundary detection, news story segmentation and key frame extraction are analyzed. All of them are the key parts of video summary. After that, three new algorithms for key frame extraction are presented, which are based on clustering algorithm with adaptive threshold, based on covariance and based on conditional entropy. The first method uses the watershed algorithm and Otsu algorithm to find the adaptive threshold. The last two method use the character that most video frames are similar to their adjacent ones to reduce the redundant information between key frames. The experimental results prove the effectiveness and efficiency of the three methods.In order to satisfy different people’s demand, a video summary method based on hierarchy is proposed, which adjusts the number of key frames according to subjective perception. At last, a COM-based video summary system is devised and implemented.更多还原