一种基于内容相似性的重复视频片段检测方法  

A detection method for near duplicate video clips based on content similarity

作　　者：刘守群[1] 朱明[1] 郑烇[1] 

机构地区：[1]中国科学技术大学自动化系,安徽合肥230027

出　　处：《中国科学技术大学学报》2010年第11期1130-1135,共6页JUSTC

基　　金：中国高技术研究发展(863)计划(2008AA01Z408;2008AA01A318)资助

摘　　要：针对互联网视频内容的复杂性,选择能够表征视频内容的特征,首先通过LSH哈希算法对特征进行索引,并由此确定视频之间的帧匹配序列,然后对于计算出的帧匹配序列,采用随机抽样一致性算法进行拟合,从而得到有效的帧匹配序列.视频之间的相似度依据有效帧匹配序列的相似度计算,由相似度进行互联网视频片段的消重.实验表明,对于大量内容相似的互联网视频片段,该方法能较好地描述相似性,并能有效提高检测准确率.In view of the complexity of video clips data,a method was designed for extracting features to represent video clips and using locality sensitive hashing to compute the frame match sequence between video clips,RANSAC was used to fit the match sequence,obtain the valid match sequence and compute the similarity between clips.Experimental results show that the method can effectively describe the similarity of web video clips as well as detect and eliminate near duplicate video clips.

关 键 词：视频片段 消重 哈希算法 LSH SIFT RANSAC 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]