Detection of video copies based on robust descriptors

In this paper, robust descriptors are extracted to detect video copies generated by complicated transformations. The main contribution of the proposed method lies in three aspects. Firstly, the complicated transformations on video copies are identified and tackled to guarantee the extraction of robust descriptors. Secondly, a motion classification approach is proposed to divide the video into video groups. Thirdly, a two-stage matching scheme is implemented and the support vector machine is utilized to detect video copies. Extensive experiments are carried out using the data from TRECVID 2008 content-based video copy detection task. The proposed framework for video copy detection is very effective, and robust against spatial and temporal variations, in comparison with TRECVID participants and state-of-art algorithms.