Video Copy-move Detection Based On Space-temporal Characteristic

With the development of information technology,the Internet has become more popular,which makes it easier to get multimedia resources.The improving level of video editing decreased the cost of manipulating videos,as which forged videos can be seen everywhere.In order to solve the problem,the government should improve the supervision,but even more important is effectively detecting videos spreading on the internet.So,for society security and theoretical study concerns,it's significant to research the algorithm to detect forged videos.There are many methods to tamper digital videos,copy-move forgery is one of the best way which can change the content and meanings of the video.The attacker can directly add or hide objects from videos,which is hard for our eyes to find the trace among them.However,there are some ploblems among this domain–simply replanting image forgery detection algorithm and weak robustness for compressing.This paper proposes two algorithms to solve these based on video region characteristics.Firstly,this paper propose a copy-move forgrey detection algorithm based on temporal shifting feature,which creatively makes use of temporal feature of videos.By extracting videos region temporal features,combining with temporal shifting characteristic,this paper utilizes phase correlation algorithm to reveal the trace of Copy-Move forgery.This algorithm first calculates region LK pyramid optical flow as temporal feature and histogram of gradient as space texture feature.With the phase correlation detection process,the algorithm then locates the forgery using the features above.The proposed algorithm creatively discovered temporal infomations hidden behind Copy-Move,and improve the performance of detection accuracy without increasing complexity.Secondly,this paper proposes a detection algorithm based on 3D-HOG feature.Considering the manipulation,no matter the faker hides the foreground through by copying from the background or inserts object from other frames,there will be intra-frame region copying.The algorithm first preprocessing the video with image threshold.Then feature matrix will be established by calculating the dense 3D-HOG feature of videos,followed by generating criterion matrix by the angle between two descriptor.Finally the algorithm locates the forgery with a sliding window.The experiment results show that this algorithm successfully utilized 3D-HOG feature to detect the footprint of Copy-Move.