基于二叉树和随机邻域嵌入的视频指纹算法

A Video Fingerprinting Algorithm Based on Binary Tree and Stochastic Neighbor Embedding

针对互联网视频认证问题,提出了一种基于二叉树和随机邻域嵌入的视频指纹算法,该算法首先利用二叉树和归一分割理论选取视频代表帧,然后以代表帧亮度的离散余弦变换系数作为视频在高维空间的特征,利用随机邻域嵌入方法把视频高维特征投影到三维空间,以视频在三维空间中相邻投影点距离序列的均值和方差构造匹配密钥,以距离序列向量二值量化构造视频指纹;对于视频指纹匹配,首先利用匹配密钥进行初级匹配,缩小匹配范围,然后在候选指纹组内进一步匹配,从而达到视频认证的目的.

In order to identify video copies on the Internet ,a video fingerprinting algorithm based on binary tree and stochas-tic neighbor embedding is proposed in this paper .In this scheme ,representative frames are selected based on binary tree and normal cut ,and the Discrete Cosine Transformation （DCT ） coefficients of luminance of representative frames are taken as the high dimen-sional features of the video .The features are mapped into three-dimensional space using stochastic neighbor embedding .A matching key is generated based on the mean and variance of distance vector between adjacent points in this three-dimensional space ,and the video fingerprint is generated by binarizing .Moreover ,during video fingerprint matching ,the matching key is used in the first-stage matching to reduce the search range ,and then a further matching is carried out in the candidate fingerprint to identify video copies .