新闻视频中非新闻段的去除  

Removing Non-news Parts from News Video

作　　者：刘宇驰[1] 栾悉道[1] 谢毓湘[1] 吴玲达[1] 

机构地区：[1]国防科技大学信息系统与管理学院

出　　处：《小型微型计算机系统》2007年第10期1837-1841,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60473117)资助

摘　　要：新闻视频研究与服务要求每天从各种电视台采集大量的新闻节目.为了节省存储空间和方便分析,要先去除新闻视频中的非新闻段(商业广告、频道/节目标识等).本文采用探测视频流中重复镜头的方法来进行此项工作.首先将新闻视频分割为镜头,再对镜头进行聚类,选择出候选的非新闻镜头聚类,然后对候选镜头进行相同镜头匹配,匹配成功的镜头即为重复的非新闻镜头.相比为其他类似问题的研究(主要是探测商业广告),该方法计算量小,不需要先验假定条件,而且实验结果令人满意.Research and service of news video demand recording volumes of news programs from many TV stations. For saving storage space and for the convenience of analysis, non-news parts （e. g. commercial and channel/program logo） of news video should be removed firstly. This paper removes non-news parts based on detecting repeating shots. First, news video is segmented into shots. Then, shots are clustered and some shot clusters are chosen as candidates of non-news shots. Finally, candidate shots are compared between each other and matched shots are removed as non-news shots. Compared with other similar study （mainly on commercial detecting）, this method needs much less computation and no prior assumption. Experimental test showed this method is satisfying.

关 键 词：新闻视频 非新闻段 镜头聚类 镜头匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]