融合局部特征和全局特征的视频拷贝检测  

Video copy detection based on local and global features

作　　者：王晶[1] 王昊[1] 

机构地区：[1]国家计算机网络应急技术处理协调中心,北京100029

出　　处：《清华大学学报（自然科学版）》2016年第3期269-272,共4页Journal of Tsinghua University(Science and Technology)

摘　　要：近年来,出于保护版权以及避免资源浪费等需要,基于内容的视频拷贝检测技术成为了一大研究热点。为了克服局部特征和全局特征的缺陷,该文提出了融合局部特征和全局特征的视频拷贝检测算法。该算法结合了分块Harris角点和分块亮度顺序,避免了采用单一特征鲁棒性差的问题,能够适应视频图像的重编码、亮度、局部遮挡和水平翻转等变化。此外,该文提出了基于信息熵的关键帧提取算法,优化了关键帧的选取;同时结合倒排索引的技术,极大地提高了检索效率。实验表明:该算法能有效地提高视频拷贝检测的准确性,同时在速度上也满足应用需要。The need for copyright protection and the desire to protect intellectual property has made content based video copy detection a widely studied issue.A video copy detection method based on local and global features is given here that overcomes the limitations of using the global and local features.The system combines blocked Harris corner and ordinal measures and is robust to illumination changes,video coding changes,partial occlusions and other effects.The system also uses inverted indexes,so this method is very efficient.Tests show that this algorithm effectively improves the accuracy of the video copy detection.

关 键 词：视频拷贝检测 HARRIS角点 分块亮度顺序(OM) 倒排索引 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]