【摘要】 为了有效地利用视频的时域运动信息来提高视频拷贝检测的精度和鲁棒性,提出一种基于特征点轨迹行为模式的拷贝检测算法.首先从视频连续帧中提取特征点轨迹的行为模式特征,然后采用视觉关键词典技术构造视频的运动特征,最后基于运动特征的相似度进行视频拷贝检测.该算法在TRECVID标准数据集上取得了较高的检测精度.实验分析表明,基于轨迹的运动特征具有较强的描述区分能力,对各种常见的拷贝变化具有鲁棒性.更多还原

【Abstract】 In order to effectively extract temporal motion information of videos for robust and precise video copy detection,this paper proposes a novel video feature based on behavior patterns of keypoint trajectories.By the algorithm,first the keypoint trajectories are generated and modeled into behavior patterns.Then patterns are learned using bag-of-visual-word method as video features.Finally,similarity between features is used for video copy detection.Comparison experiments on benchmark video dataset(TRECVID)show that proposed algorithm achieves high detection precision.And this work demonstrates that the trajectory based feature has better robustness for various visual changes.更多还原