Research Of Key Technology On Content-Based Video Copy Detection

With the rapid development of multimedia technology and the spreading of network communication technology, digital media are widely used, and the video has been gradually one of the main ways for the information dissemination. Video can be converted into a variety of different versions by multi-media processing tools. So the video providers have been challenged by the digital copyright protection. Content-based Video copy detection (CBCD), is emerging as an alternative to the traditional watermarking approach to cope with the digital video piracy and illegal distribution problems. In this paper, the technology has been researched deeply, and several new algorithms have been proposed. Key problems in CBCD have been solved in three aspects:the representation of visual content, the retrieval and matching of videos, and the index of video features.In the area of the representation of visual content, first the algorithm of a key point selection is optimized, which optimize the local feature "SIFT", which is invariant to rotation and translation. Then, the high-dimensional features are quantified by the algorithm of points' projection based on the visual vocabulary. At last, the binary signature and the ordinal signature are proposed to improve the matching precision. Experimental results show that the algorithm can effectively balance the feature points matching speed and accuracy, and the feature meets the needs of the video copy detection very well.In the aspect of the retrieval and matching of videos, a matching strategy based on the continuity in time for video frames is proposed, and the detection accuracy is improved. Group the matching frames by the difference of frames. Then the candidate matching video segments are refined, which aims to get the optimal results. Experimental results indicate that the false detection rate has been effectively reduced.In the respect of the index of video features, an inverted structure based on the visual vocabulary is designed, by which the retrieval speed is raised. Experimental results verify that the detection efficiency is deeply increased and the fast query in a huge number of videos is achieved.In short, a common system framework is proposed for video copy detection. The evaluation results of TRECVID 2009 show that, the performance of the system is much higher than the average level of all teams which have participated in the competition.