闪光灯和标题条对新闻视频镜头检测影响的研究  

News video shot detection analysis for closed-captions and flash

作　　者：张子银[1] 白雪生[1] 徐光祐[1] 鄢恒权[1] 

机构地区：[1]清华大学计算机科学与技术系,北京100084

出　　处：《清华大学学报（自然科学版）》2003年第1期71-74,共4页Journal of Tsinghua University(Science and Technology)

基　　金：国家"八六三"高技术项目(863-306-ZT04-02-01及863-306-QN99-01);国家自然科学基金资助项目(60273005);高等学校博士点专项科研基金资助项目(99000307)

摘　　要：新闻视频中特有地存在着大量闪光灯事件和后期视频编辑时加入的标题条事件,对镜头检测带来很大干扰。开发了一种建立在统计学基础上的稳健帧间差度量算法,克服标题条的突然出现和消失等局部剧烈运动对镜头切分的影响。根据闪光灯在视频流时序上的特征,提出通过对潜在切变位置前后视频内容的再检测来区分闪光灯事件和切变。该方法具有通用性和实时性,可以在视频自动分析或自动索引系统中得到实际应用。The large amount of flash and closedcaption events seriously hampers shot detection in news videos. A robust dimension was developed to overlook the differences between two consecutive frames due to local acute movements, such as the emergence and disappearance of closedcaptions. Comparison of the relative content of two images (one before and the other after the potential cut position) can identify a flash event which would then be labelled as a cut event. These methods are both generic and realtime, and can be used in automatic video analysis and automatic indexing systems.

关 键 词：新闻视频 镜头检测 视频分析 切变检测 闪光灯 标题条 统计学 视频结构 镜头切分 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]