基于内容的新闻镜头分类技术研究

发布时间：2018-02-17 05:37

本文关键词： 新闻视频镜头边界检测镜头分类广告检测条件随机场　出处：《天津大学》2008年硕士论文　论文类型：学位论文

【摘要】： 随着网络和计算机技术的高速发展,多媒体信息,特别是数字视频越来越多的进入人们的生活。如何对视频信息有效管理和利用,是摆在人们面前的重要课题。因此,在理解视频内容的基础上,建立视频的索引、浏览和检索等应用系统,提供给用户方便的视频内容获取方式就成为研究人员努力的方向。镜头作为视频中的结构单元,既可分割成为图像帧,也可组合构造成故事单元,因此在视频内容分析技术中,对镜头进行分析具有十分重要的作用。对视频中的镜头进行有效的分类,一方面,可以极大地缩短低层视觉特征与高层语义特征之间的“语义鸿沟”;另一方面,镜头分类还是视频摘要、索引、检索等视频管理和应用技术重要的支持和保证,具有十分重要的现实意义。新闻视频是内容结构性比较强的视频类型,本文针对新闻视频,设计了一种基于内容的镜头分类方法。该方法将新闻视频中的镜头分为主持人、记者、独白、广告、静态图像以及“其他”六个类型。其中,“其他”指新闻视频中除去另五类镜头后剩余的镜头。主持人、广告、静态图像和“其他”这四类镜头,根据其自身特点逐一检测。记者和独白镜头在新闻视频中是最难以区分的,为此,本文利用一种机器学习方法——条件随机场,将记者和独白镜头的分类转化为序列标注问题,并进行了实验,得到了不错的效果。
[Abstract]:With the rapid development of network and computer technology, multimedia information, especially digital video, is coming into people's life more and more. How to manage and utilize video information effectively is an important subject in front of people. Based on the understanding of video content, the establishment of video indexing, browsing, retrieval and other application systems to provide users with convenient access to video content has become the direction of researchers. As the structural unit of video, the shot can be divided into image frames or combined into story units, so in video content analysis technology, It is very important to analyze the shot. On the one hand, the "semantic gap" between the low-level visual feature and the high-level semantic feature can be greatly shortened by effectively classifying the shot in the video; on the other hand, the "semantic gap" between the low-level visual feature and the high-level semantic feature can be greatly reduced. Shot classification is also an important support and guarantee of video management and application technology, such as video abstract, index and retrieval, which has very important practical significance. News video is a kind of video with strong content structure. This paper designs a content-based shot classification method for news video. This method divides the shot of news video into host, reporter, monologue, advertisement, etc. "other" refers to the rest of the shots in news videos after the other five shots are removed. The four types of shots are the anchorman, the advertisement, the still image, and the "other". In this paper, a machine learning method, conditional random field, is used to transform the classification of journalists and monologues into sequence tagging problems. Experiments are carried out and good results are obtained.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2008
【分类号】：TP391.41

【参考文献】