基于视频指纹的视频片段检索方法.pdfVIP

第36 卷 第7 期 计 算 机 工 程 2010 年4 月 V Computer Engineering April 2010 ol.36 No.7 ·多媒体技术及应用· 文章编号：1000—3428(2010)07—0239—03 文献标识码：A 中图分类号：TP37 基于视频指纹的视频片段检索方法 1,2 1,2 1,2 1,2 李泽洲 ，欧阳建权 ，张 敏 ，刘 炜 (1. 湘潭大学智能计算与信息处理教育部重点实验室，湘潭 411105 ；2. 湘潭大学信息工程学院，湘潭 411105) 摘 要：提出一种基于视频指纹的视频片段检索方法，通过视频指纹的相似性匹配对视频片段进行检索。通过改进的Harris 检测计算DC 图像的角点信息，去掉视频图像边缘提取视频运动特征，根据图像特征相似度和运动特征相似度的调和因子得到视频片段总的相似度。实 验结果表明，与现有方法相比，该方法能准确检测视频片段，具有较强的鲁棒性，满足视频检索的实时要求。 关键词：视频指纹；Harris 检测；视频检索 Video Segment Retrieval Method Based on Video Fingerprint LI Ze-zhou1,2, OUYANG Jian-quan1,2, ZHANG Min1,2, LIU Wei1,2 (1. Key Laboratory of Intelligent Computing Information Processing, Ministry of Education, Xiangtan University, Xiangtan 411105; 2. College of Information Engineering, Xiangtan University, Xiangtan 411105) 【Abstract 】This paper produces the fingerprint by combining both the image feature based on the improved Harris detection and the motion feature based on the differences between the adjacent video frames, and retrieval video segment through similarity match of the video fingerprint. The corner descriptor of the DC image is generated based on the improved Harris method. Moreover, the motion feature is extracted by trimming the image edge of the video. The fusion factor which combines the similarity of image feature and motion feature is constructed to calculate the similarity between the candidate videos and original videos. Compared with existing method, the method proposed in this paper retrie