The Research On The Algorithm Of Content-Based Video Copy Detection

Along with the development of computer, web, and other multimedia technologies, the amount of videos increases exponentially nowadays. The spread of the videos leads to many copies on the Internet or TV. Therefore, a video copy detection system is necessary for the digital copyright protection, information(illegal content) monitoring, video retrieval and so on.In this paper, a video copy detection system which can effectively address these problems is presented, and also two kinds of local features for a subsample of frames from the video are extracted. One is the 128-d SIFT descriptor, which is employed to cope with spatial content-altering transformations. The other is the 32-d Oriented Features from Accelerated SegmentTest(FAST) and Rotated BRIEF(ORB) descriptor. It is a very fast binary descriptor based on Binary Robust Independent Elementary Features(BRIEF), which is rotation invariant and resistant to noise. These two descriptors are combined together, achieving total robustness to various transformations. In order to reduce dimensionality and index efficiently, the local descriptors are hierarchically quantized in a vocabulary tree. This vocabulary tree allows a larger and more discriminative vocabulary to be used efficiently. Each keyframe is represented by a vocabulary histogram. The entire set of the representations of the video dataset is stored in a structure similar into the inverted file, which accelerates feature matching process. For video copy detection, three sequence matching methods are proposed. Two of them are based on the histogram, and the other is based on diagonal. The experimental results on the TRECVID 2011 video copy detection dataset show that the proposed system is effective and efficient.