Chinese Journal of Computers   Full Text
Title	Spatiotemporal Video Copy Detection Based on Visual Perception Analyses
Authors	PAN Xue-Feng1),2) LI Jin-Tao1) ZHANG Yong-Dong1) TANG Sheng1) YU Le-Jun1),3) XIA Tian1),2)
Address	1)(Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190) 2)(Graduate University of Chinese Academy of Sciences, Beijing 100190) 3)(Department of Electronic Information Science and Technology, Beijing Normal University, Beijing 100875)
Year	2009
Issue	No.1(107—114)
Abstract &   Background	Abstract A novel spatiotemporal video copy detection method based on visual perception features is proposed in this paper. The method differs from the conventional video copy detection algorithms in that it does not using key frame based features any more. By extracting visual perception based spatiotemporal correlation of slice in frequency domain using DCT, the proposed method possesses both robustness and discriminability for video copy detection. It copes well with not only the format, resolution variation of video but also the aspect ratio change. Even more, the method is robust to display format conversion such as adding pillar-box. The experiment results show the proposed method is effective and robust. Keywords video copy detection; visual perception; spatiotemporal correlation; spatiotemporal slice; DCT Background More and more video contents are encoded in different formats to meet different requirements. How to find the video content in different formats is important for video retrieve and search. Some format-independent video copy detection methods were proposed. The ordinal measure (OM) has shown superior performance than motion and color based features for format-independent video content representing. According to visual perception theory, the visual sensitivity related motion information is very important to human visual perception. And it will be remained stably in encoding process. So a format-independent motion describing method based on visual sensitivity is proposed in this paper. The spatiotemporal slice DCT is used to extract the effective features for video content. Experiments show the proposed describing method is sensitive to variation of video content and is robust to variation of video format and the method proposed performance well in video copy detection. The work in this paper is partially supported by the National Basic Research Program (973 Program) of China (2007CB311100), the National High Technology and Research Development Program(863 Program) of China (2007AA01Z416), the National Nature Science Foundation of China (60773056), and the Beijing New Star Project on Science & Technology (2007B071).