A Secure Client Video Deduplication Scheme Based on 3D CNN

With the advent of the era of big data, users are generally used to uploading local data to the cloud server to reduce the pressure of storage. This will lead to many duplicate data in the cloud server and form a large number of redundant, especially the existence of videos. To solve the problem of wasting storage space for duplicate videos, on the one hand, we propose a two-stream 3D CNN combined with triple metric method to construct a duplicate video detection scheme, when the threshold is 3.6, the recall rate can reach 97.1905%, while the false positive rate is only 0.0743%, which is better than the I3D deduplication scheme. Through the Chamfer distance matching algorithm for two-stream video features, compared with the duplicate video detection model based on I3D network with EMD distance in terms of false positive rate, recall rate and time cost, it achieved a good video deduplication effect. In addition, some videos are used to test the deduplication performance of Hamming distance, EMD distance and Chamfer distance, the results show that the Chamfer similarity distance algorithm has a higher recall rate and a lower false positive rate than the other two methods. On the other hand, we propose a video ownership proof scheme based on random block fusion. The server and the client respectively perform random cropping operations and fusion on the video frames, then generate verification tags and ownership evidence. Finally the server calculates the distance to determine whether the client can pass the authentication. The experiment results show that the scheme has less time and storage overhead and can resist the spoofing attack of criminals.