Research And Application Of Copy Detection Based On Video Fingerprint

With the development of the mobile internet and the popularity of smartphones,video has become an important medium for information dissemination,and at the same time has given rise to a large number of infringements against copyrighted videos.With the massive amount of data being uploaded on all major video sharing sites today,the use of manual means for video auditing has become overwhelming.In order to protect the rights of video production companies and original producers,there is an urgent need to detect infringement against short videos through automated methods.Due to the scale and diversity of infringements,content-based video copyright detection is a challenge.Traditional video copy detection techniques cannot cope with the wide variety of video infringement methods.To solve this problem.This paper designs an accurate and efficient video copyright detection model based on video fingerprint matching technology;it is able to automatically determine whether a video to be reviewed is infringing and locate its position in the copyrighted video based on the library of copyrighted videos provided and the video to be reviewed.The main work of this paper is as follows.(1)decoding the copyrighted video and the video to be examined,extracting the key frame feature vector,adopting the unsupervised aggregation feature extraction method,and proposing Gaussion_R-MAC based on the convolutional layer of neural network to achieve video fingerprint extraction;(2)performing nearest neighbour search on the feature vector by NSG algorithm,and proposing a nearest neighbour search algorithm combined with the idea of re-ranking;(3)Matching the nearest neighbour feature vector with the reference set feature vector through the video matching algorithm to obtain the final matching result;(4)Implementing a platform for automated copyright detection on the basis of the video copyright detection model,providing a concise and clear interactive interface for users.Through comparison experiments with existing algorithmic models,it is proved that the model in this paper can achieve better results in the detection and localization work of copyrighted videos.At the same time,the platform allows users to easily perform video detection online.