融合多特征的新闻故事探测  

News Story Detection Method Based on Multi-feature Fusion

作　　者：栾悉道[1] 谢毓湘[1] 刘宇驰[1] 吴玲达[1] 刘洋[2] 

机构地区：[1]国防科技大学多媒体研究开发中心,湖南长沙410073 [2]北京图形研究所,北京100029

出　　处：《小型微型计算机系统》2008年第5期950-953,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60473117)资助;国家“八六三”高技术项目基金项目(2006AA01Z319)资助;国防科技大学优秀研究生创新资助

摘　　要：探测出新闻故事的边界,将对新闻视频的结构化分析以及新闻视频的语义理解产生积极的作用.通过对新闻视频的结构进行分析,提出了一种融合镜头、播音员、标题字幕、静音等多种特征的新闻故事探测方法.该方法避免了依赖单特征探测新闻故事边界的不足,实验证明,该方法可以获得平均96%的故事边界探准率,能够较好地完成新闻故事边界探测的任务.It is important to analyze the structures of news videos and to understand their semantic content,news story boundary detection is one effective way of solving the above problems. By the preliminary analysis of news structures ,a news story detection method based on multi-feature fusion is proposed. The method avoids the deficiency of detecting story boundaries by a single feature, but fuses multi-features such as shots,anchorpersons ,titles and silences ,etc. Experiments prove the effectiveness of the method,which can get the average detection precision of 96% ,and can accomplish the task of detecting story boundaries.

关 键 词：新闻视频 故事探测 多特征融合 播音员镜头 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]