首页 | 本学科首页   官方微博 | 高级检索  
     

基于音频的视频拷贝检测
引用本文:赵花婷,;王明敏.基于音频的视频拷贝检测[J].微型电脑应用,2014(7):1-4.
作者姓名:赵花婷  ;王明敏
作者单位:[1]复旦大学计算机科学与技术学院; [2]东方有线网络有限公司
基金项目:国家863项目(2011AA01A100);上海市科委项目(12dz1500203,12511505303)
摘    要:提出了一个兼具实时性和鲁棒性的短视频拷贝检测框架,并以长视频中的广告检测为应用进行研究.框架主要分为两个部分:1.基于音频匹配的短视频拷贝检测;2.基于图像的边界精确定位.第一阶段从原始视频中获得声谱图,并进一步地提取出音频特征并在音频库中寻找匹配;在第二阶段,对检测到的视频片段进行合并,然后利用视频片段在边界处附近的图像特征,对边界片段进行精确定位.实验结果显示,准确率高达99%,取得了很好的效果.

关 键 词:视频拷贝检测  短视频  音频匹配  平滑  静音段  图像特征  实时

Audio-Based Duplicate Video Detection
Affiliation:Zhao Huating,Wang Mingmin( 1.Department of Computer Science, Fudan University, Shanghai 201203, China; 2.Shanghai OCN CO.Ltd., Shanghai 201203, China;)
Abstract:This paper proposed an algorithm framework,which is both real-time and robust,for short video copy detection,and researched the application in terms of long video commercials detection.The framework was consisted of two parts:1.Audio-Based short video copy detection; 2.Vision-based precise video border location.Stage 1 of study,which extracted spectrogram from original video,and further extracted audio features,then searched for the match in the sample library.In stage 2,it merged candidate video clips,then precisely pinpointed the video border using image feature along with the candidate border area.Experimental results show accuracy rate of up to 99%,has yielded very good results.
Keywords:Video Copy Detection  Short Video  Audio-Match  Smooth  Silent Segment  Image Features  Real Time
本文献已被 维普 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号