【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。更多还原

【Abstract】 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.更多还原