文章摘要
基于深度学习的视频指纹快速鲁棒匹配方法
Fast robust video fingerprint matching method based on deep learning
投稿时间:2022-07-14  修订日期:2022-07-14
DOI:
中文关键词: 深度学习  视频指纹  分层匹配  鲁棒性  特征点  极坐标系  
英文关键词: deep learning  Video fingerprint  Hierarchical matching  Robustness  Characteristic points  Polar coordinate system  
基金项目:2021安徽省自然科学研究重点项目(KJ2021A1384)
作者单位邮编
李超 铜陵职业技术学院 信息工程系 244000
摘要点击次数: 42
全文下载次数: 0
中文摘要:
      视频指纹匹配时主要通过手工特征方法获取表层特征信息,使得匹配结果的F-value值较低。为此,依托于深度学习网络,设计视频指纹快速鲁棒匹配方法。针对给定视频提取梯度方向质心,获取待匹配视频指纹。应用卷积神经网络,构建深度学习特征提取模型,得到视频指纹二值化特征,并通过平移、缩放等几何操作变换指纹细节特征。引入分层匹配理念,通过包含粗匹配和精匹配两个环节,实现视频指纹快速鲁棒匹配。实验结果显示:视频未编辑和亮度参数编辑后,所提方法的平均F-value最高值为0.95,匹配效果较好。
英文摘要:
      In video fingerprint matching, the surface feature information is mainly obtained by manual feature method, which makes the f-value of the matching result low. Therefore, based on the deep learning network, a fast and robust video fingerprint matching method is designed. Extract the gradient direction centroid for a given video, and obtain the video fingerprint to be matched. Using convolution neural network, a deep learning feature extraction model is constructed to obtain the binary features of video fingerprint, and the detailed features of fingerprint are transformed through geometric operations such as translation and scaling. The concept of hierarchical matching is introduced to realize fast and robust video fingerprint matching by including two links: coarse matching and fine matching. The experimental results show that the maximum average f-value of the proposed method is 0.95 after the video is not edited and the brightness parameter is edited, and the matching effect is good.
View Fulltext   查看/发表评论  下载PDF阅读器
关闭

手机扫一扫看 分享按钮