Research On Content-based Video Copy Detection

The rapid development of digital media and network technology makes huge amounts of copied videos emerging.These bring difficulties in video storage,video copyright protection,video retrieval and etc.As an effective solution to the problem,contented-based video copy detection has become a research focus in the field of multimedia information processing.This thesis studies content-based video copy detection,in which three kinds of new methods and improvement measures are put forward to improve the efficiency and the accuracy of the copy detection system.Main contributions of the thesis are summarized as follows:To improve the operating efficiency of video copy detection system,an algorithm fusing color and gradient histograms is proposed,which could be used in shot segmentation in video copy detection.Meanwhile,an algorithm based on color histogram cumulative difference is proposed in key frame extraction.This thesis presents a new algorithm based on global feature and local feature.HSV color histogram and gradient histogram are used to describe the global feature.Meanwhile,the thesis studies the performance of five kinds of local feature extraction algorithms(SIFT?SURF?ORB?BRISK?AKAZE)from different standpoints,and utilizes AKAZE feature extraction algorithm to describe the local feature because of its excellent comprehensive performance.The experimental results show that the algorithm yields the recall rate of 96.80%and the precision rate of 97.95%,but still with the shortage of the high time complexity.To improve the time performance of video copy detection system,this thesis presents a fast video copy detection algorithm by improving the traditional bag of words.The algorithm utilizes GIST feature and PHOG feature to describe the global feature,and utilizes bags of video words generated by clustering analysis to describe the video and subject to matching.The experimental results show that the algorithm yields the recall rate of 87.8%and the precision rate of 100%,and improves the detection efficiency apparently.In addition,this thesis presents a video copy detection algorithm in compression domain.Firstly,I-frames of video is divided into blocks,and each block is traversed by Hilbert-order to generate the Hash value about the average of DC and AC coefficients respectively.Then it utilizes the DC and AC features to match sequentially.The experimental results show that the algorithm yields the recall rate of 88.6% and the precision rate of 87.96%,and has a higher detection efficiency.