一种基于SVM主动学习的卡通视频检测方法  

A Cartoon Video Detection Method Based on Active SVM Learning

作　　者：高新波[1] 田春娜[1] 张娜[1] 

机构地区：[1]西安电子科技大学电子工程学院,西安710071

出　　处：《电子与信息学报》2007年第6期1338-1342,共5页Journal of Electronics & Information Technology

基　　金：新世纪优秀人才支持计划(NCET-04-0948);教育部重点项目(104173);国家自然科学基金(60202004);教育部留学归国人员实验室资助课题

摘　　要：通过分析卡通与非卡通视频在视觉上的差异,对视频片断提取了MPEG-7描述子等8组视觉特征来构造卡通视频的特征空间;并将主动相关反馈技术引入到支撑向量机(SVM)算法中,设计了一种基于主动学习的卡通视频检测分类方法。利用大量实际视频片断所做的测试实验结果表明,该文选取的特征对卡通和非卡通视频有较好的区分能力;且与单纯的SVM算法以及传统相关反馈和SVM算法结合的方法相比,该文算法在检测性能上有较大的优势。Through analyzing visual differences between cartoon and other videos, 8 groups of typical features including MPEG-7 descriptors are extracted to construct the feature space of cartoon videos. Then, a content-based video classifier is designed by introducing the active relevance feedback technique into Support Vector Machine （SVM） for the cartoon video detection. Experimental results on a great many real video clips illustrate that the constructed feature space can represent the cartoon videos effectively. In addition, compared with the classifier based on SVM and that based on the traditional relevance feedback technique and SVM, the proposed classifier has a higher performance for cartoon video detection.

关 键 词：视频检测 MPEG-7 支撑向量机 相关反馈 主动相关反馈 

分 类 号：TN919.8[电子电信—通信与信息系统]