Spatio-temporal video copy detection

Video copy detection algorithms are used to find copies of original video content even if the content has been altered. Given the prevalence of video recording and copying devices as well as the availability of many Internet sites for hosting videos, detecting video copies has become an important problem especially for companies interested in managing and controlling copyrights of their content. We propose a new content-based video copy detection algorithm. The proposed algorithm creates signatures that capture the spatial and temporal features of videos. These spatio-temporal signatures enable the algorithm to provide both high precision and recall. In addition, these signatures require small storage and are easy to compute and compare. Our extensive experimental analysis with a large video dataset shows that the proposed algorithm achieves high precision and recall values while remaining robust to many video transformations that commonly occur in practice. The algorithm is simple to implement and is more computationally efficient than previous algorithms in literature.