Research Of Content Based Video Copy Detection

With the development of multimedia technology and video data explosive growth,people can easily get video data from the internet and share to others,and video data can be modified by any people with editing software easily,which result in more and more video copies appears in our lives.In daily life,people are more concerned about the content of the video,such as recording the real scene of the world and wonderful moments in life,while ignoring its problems.After a long time,the number of video data formed a wide variety of video database.How quickly and accurately retrieved video and its copies,it has become an important problem to be solved.Based on these requirements,we propose a video copy detection algorithm based on temporal contextual hashing.The proposed method mainly extracts the temporal context information of the key frames and expresses the temporal context information of key frames as binary codes,and compares the key frames' binary codes by calculating Hamming Distance to achieve temporal verification efficiently and implicitly.In order to accelerate the speed of video copy detection,we use LSH algorithm to build index for all the key frames in reference video database.Experimental results on the publicly available video database(TRECVID 2009)indicate the proposed approach achieves high efficiency and accuracy.Although the above method has a better retrieval effect on common transformations,but it is unable to have a good performance on complex copy transformation because of using block-based color correlation histogram as the features of the video frame,which result in the lower robustness of the algorithm.Therefore,based on the above method,we proposed another video copy detection algorithm based on spatial-temporal contextual code.The method combine the local feature SIFT and bag of words model with the contextual model to generate spatial-temporal context information.In spatial,we express the content of key frame by extracting spatial context information and in temporal,we extract temporal context information based on the bag of words model with SIFT.Then we represent the spatial and temporal context information into binary code.Experimental results show that this method is more effective in the complex transformation.