A robust video hashing for video copy detection based on Laplacian Eigenmaps (LE) is proposed. In this method, key frames are first selected based on video tomography and a uniform distribution vector, and then the four-order cumulants are taken as the feature of video in the high dimensional space. The video is mapped to a three-dimensional space using LE, and video hashes are generated using the norms of the points. Experimental results show that the video hashing is robust and discriminative.%针对视频拷贝检测问题,提出了基于拉普拉斯特征映射(Laplacian Eigenmaps,LE)的视频哈希方法,该方法利用视频层析成像技术和服从均匀分布的向量对视频进行镜头分割和关键帧提取,以高阶累计量作为视频在高维空间的特征,并利用LE进行降维,得到视频在三维空间中的轨迹,利用三维空间中点的范数构造视频哈希来实现视频拷贝检测.实验结果表明,该方法具有较好的鲁棒性和区分性.
展开▼