Video Copy Detection using Motion Co-occurrence Feature

Content-based video copy detection (CBCD) has an important role especially in distributing and tracking of copyright and commercial videos. Basically, two main descriptors are used for CBCD one of which, using the interest points in keyframes, the other using the content of a whole keyframe. Even if the local descriptor-based approaches have good results in CBCD problem, it takes too much time for extracting and indexing features. In this study, one of the motion-based features named as motion co-occurrence feature (MCF) is proposed for the solution of CBCD problem. Basically, MCF is a global feature that uses local information by considering spatial and temporal neighbourhood of motion vectors. Additionally, MCF can be extracted directly from the motion information of the video bitstream. Thus, feature extraction process can be performed quickly. The results show that the proposed method is effective especially for the attacks in which the motion information is preserved.