In order to identify video copies on the Internet

a video fingerprinting algorithm based on binary tree and stochastic 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 dimensional 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.