A Rotation Invariant BSIF Descriptor for Video Copy Detection Using a Ring Decomposition

Recently, with the increasing number of Internet users, detecting non legal copy of video sequence is of outstanding importance. This paper provides a new approach for content based video copy detection (CBVCD) which is invariant to rotation and flipping attacks. The proposed scheme is based on binary statistical image features (BSIF) descriptor using a new ring decomposition. The ring partition is particularly suitable for rotation/flipping attacks that affect the video frames. In fact, the visual content of each ring is kept constant when the video frames are rotated or flipped. The proposed VCD system was evaluated under TRECVID 2009 database and compared to others algorithms based on local binary pattern (LBP), local phase quantization (LPQ) or histogram of oriented gradient (HOG) descriptors. The experimental results demonstrated that the proposed descriptor is effective for all the attacks that can affect a video sequence and particularly in the case of rotation and flipping attacks.