基于AP聚类和频繁模式挖掘的视频摘要生成方法  

APPROACH FOR GENERATING VIDEO SUMMARISATION BASED ON AP CLUSTERING AND FREQUENT PATTERN MINING

作　　者：顾诤[1] 肖若贵 

机构地区：[1]浙江大学计算机科学与技术学院,浙江杭州310027 [2]杭州华星通信技术股份有限公司,浙江杭州310012

出　　处：《计算机应用与软件》2010年第6期5-7,38,共4页Computer Applications and Software

基　　金：国家自然科学基金项目(60803110)

摘　　要：为了有效支持视频数据库浏览和检索,通过视频摘要来对视频进行紧凑表达变得十分重要。提出了一种新颖的基于近邻传播聚类AP(Affinity Propagation)和频繁镜头模式挖掘的视频摘要自动生成算法。视频频繁镜头模式被定义为在一定时间窗口内经常出现的镜头系列。首先通过近邻传播聚类,将相似镜头聚合到一起;然后采用频繁镜头模式挖掘的方法对视频聚类内容进行挖掘,去掉视频中冗余内容部分;最后通过覆盖视频语义信息的频繁镜头模式生成视频摘要。实验结果表明,视频摘要算法取得了良好的效果。In order to effectively support video database browsing and retrieval,it is essential to compactly represent the video by video summarisation.In this paper,we present a novel automatic generation algorithm of video summarisation based on affinity propagation（AP） clustering and frequent shot pattern mining.The video frequent shot pattern is defined as a series of video shots which occur frequently within certain time slot window.First,we cluster similar video shots through affinity propagation clustering,and then mine the contents of video clustering by means of frequent shot pattern mining to remove visual-content redundancies.Finally the video summarisation is generated by those frequent shot patterns with time interval constraint to cover the video scenario information.Experimental results indicate that the proposed algorithm achieves quite good effect.

关 键 词：近邻传播聚类 频繁镜头模式 视频摘要 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]