【作者】 任绍美； 都云程； 王涛； 吕学强；

【Author】 REN Shao-mei1,DU Yun-cheng1,2,WANG Tao1,2,LV Xue-Qiang1,2 1Chinese Information Processing Research Center,Beijing Information Science and Technology University,Beijing,China,100101 2Beijing TRS Information Technology Co.,Ltd.Beijing,China,100101

【机构】 北京信息科技大学中文信息处理研究中心； 北京拓尔思信息技术股份有限公司；

【摘要】 场景分割是基于内容的视频索引和检索研究的重要内容。本文采用基于核密度估计的均值漂移算法求出每个子镜头对应颜色特征的核密度极大值,该算法时空复杂度低,而且场景的代表帧也可根据此结果计算得出。本文提出一种自适应方法实现场景的自动分割,该方法既不需要进行训练也不需要给出场景类别数等先验知识。并且该方法的提出不依赖于视频的特定领域知识,所以其适用范围广。对比实验结果表明,视频场景分割的查全率和查准率都好于参考文献的方法。更多还原

【Abstract】 Scene segmentation is an important research part for the Content-Based Video Indexing and Retrieval.The mean shift algorithm based on the kernel density estimation is used to calculate the maxima point of the nuclear density of the sub-shots’ color feature.The time and space complexity of the algorithm is relatively low,and the key-frame of the scenes can be easily calculated from those points.An Adaptive Approach,which does not need to train data set and give the priori knowledge of the number of scene categories,is proposed to split scenes automatically.And more,this method is applicable to many types of video because of without depending on the video domain-specific knoledge.The experiment results shows that the recall and precision rates is obviously better than the reference literatures’.更多还原