A relative phases algorithm of key edges for detecting compressed video copies

A new algorithm to detect compressed video copies is proposed to improve the recall and speed of video copy detection based on the relative phases of key edges. The algorithm firstly computes edge information from the AC components of the DCT coefficients, which are directly obtained from all the I frames in compressed videos; The relative phases of key edges are computed from the edge information, and then used as the video signatures; Finally the video signatures are used to detect the video copies. Since various video copy attacks have little influence on the relative phases of key edges, the proposed algorithm is robust. Moreover, the algorithm avoids the decompression process so that the detection time is reduced. Experimental results show that the proposed algorithm outperforms not only the other compressed-domain algorithms 68% in recall, but also the uncompressed-domain algorithm about 5 times in speed.