【作者】 谢林江；

【导师】 包秀国；

【作者基本信息】 北京邮电大学 ， 电子与通信工程（专业学位）， 2013， 硕士

【摘要】 网络技术发展日新月异,多媒体技术应用领域也日益广阔。如何快速的浏览视频内容,如何简洁的表达视频的主要内容,以及如何实现对大容量视频库进行快速检索等方面的问题已经成为研究的热点。视频摘要技术在这种背景下应运而生,成为解决问题的有效手段。分析近年来国内外视频摘要相关文献,可以发现摘要关注的角度以及使用的方法都会影响视频摘要系统的处理效果。本文将研究和实现一种基于内容的视频摘要系统。针对镜头边界检测中遇到的问题,本文提出了一种基于图像局部特征点匹配的镜头边界检测算法。该算法采用SURF算子进行图像局部特征检测,分为粗检测和细检测两个阶段进行镜头边界检测。其中粗检测目的是推荐可能存在边界的区间,而细检测的目的是判断推荐区间里是否存在镜头边界以及确定边界位置。在一系列测试实验中,该算法表现出了较好的稳定性以及准确性。针对镜头聚类的问题,本文将近邻传播聚类算法应用于镜头聚类。实验结果证明该方法有效而且合理。在设计镜头重要度模型过程中,针对图像背景比较复杂这一情况,本文提出一种基于多特征融合的人脸检测算法。通过对比试验以及在线检测实验,证明了该算法具有良好的稳定性以及准确性。为了实现视频摘要,以镜头重要度模型为基础设计了一种视频摘要生成方法。最后,本文结合研究成果设计和实现了一个基于内容的视频摘要系统。该系统采用C++语言,OpenCV开源库以及其它相关库实现,具备一定的应用价值。更多还原

【Abstract】 As the rapid development of network technology, the application fields of multimedia technology are also increasing broadly. It has become a hot research topic that how to browse video content quickly, how to express the main content of the video succinctly as well as how to achieve fast retrieval of large-capacity video library and so on.In this context, the video abstract technology emerged as an effective method to solve the problems. The analysis of the domestic and foreign relevant literature about video abstract in recent years is that the perspective of video abstract concern about and methods used by video abstract have affected the treatment effect of video abstract system. This dissertation would research and implement a content-based video abstract system.With some questions met in the shot boundary detection(SBD), this dissertation proposed a algorithm for SBD based on the local image feature points matching. local Image features are detected by using the detector of SURF in the algorithm and there are two phases of SBD, that is, rough detection and fine detection. The purpose of the rough detection recommended the possible existence of boundary interval and the fine detection is to judge whether there is shot boundary in the recommended boundary interval as well as to determine the location of the boundary. In a series of test experiments, the algorithm shows good stability and accuracy. In term of shot clustering, the affinity propagation clustering algorithm is applied to the shot clustering. Experimental results show that the method is effective and reasonable. In terms of the image with complex background during the design of shot important degree model, this dissertation proposed a face detection algorithm based on multi-feature fusion. Through the contrast test as well as online detection experiments, it proves that the algorithm has good stability and accuracy. The video abstract is created by a video abstract generation method based on shot important degree model. Finally, according to the research achievement, this dissertation designed and implemented a system of content-based video abstract. The system was realized by using the C++language, OpenCV open source library and other related libraries, with a certain application value.更多还原