Research On Forensics Algorithms Of Digital Video Inter-frame Forgery

As a carrier of information,digital video plays an important role in daily life.Due to the rapid development of video editing tools,people can modify video content more easily,which leads to a huge challenge to its authenticity.In some major emergencies,once the falsified videos are uploaded to public networks or used as exhibits,they may threaten national security,social stability and normal life order.Therefore,it is necessary to study fast and effective video forensics methods.Since inter-frame tampering is the most probable forgery method due to its simple operation.Therefore,there is an urgent need to detect inter-frame forgery effectively.Researchers have conducted extensive research on traditional interframe forgery such as frame deletion,frame duplication and frame insertion,and successful forensic algorithms are available.However,there is a lack of effective detection methods for new types of forgery such as video speed manipulation and video frame inversion.Therefore,this paper focuses on the new interframe forgery methods and proposes forensic solutions.After an in-depth analysis of the latest video inter-frame forgery detection schemes,three detection algorithms are proposed in this paper,and the main work includes:(1)A slow video detection algorithm based on temporal feature representation is proposed.Since slow videos are faked by uniformly introducing interpolated frames,the correlation between video frames will show periodic changes,so the algorithm calculates the autoregressive model coefficients of the pixel sum of frame differences as time-domain features firstly.Then in order to further explore the interpolated traces on the spatial domain,Markov features based on frame differences are introduced.Next the two features are combined to obtain the joint spatio-temporal features.Finally,the integrated classifier is used to achieve video authenticity discrimination.The experimental results show that the scheme improves the detection accuracy under the two slow forgery operations effectively.(2)A fast video detection algorithm based on frame encoding macroblocks and pixel change features is proposed.The forgery principle of fast video is to delete frames uniformly,which will lead to a larger prediction error when inter-frame coding is performed.Since the prediction error becomes larger,the number of different types of macroblocks changes,so the algorithm extracts the sequence of different macroblocks changes as the object of study to obtain the periodic features based on the number of macroblocks changes firstly.Then in order to make full use of the inter-frame correlation information brought by frame deletion,the periodic features of pixels and changes of frame difference are extracted.Finally,the fused features are further filtered out from the noise interference using the autocorrelation function.The features are classified by support vector machine.The experimental results show that the algorithm can improve the accuracy rate when the forgery traces are weak significantly.(3)A video frame duplication with shuffling detection algorithm based on local binary pattern feature matching in spatio-temporal domain is proposed.Frame duplication with shuffling is to copy the partial frames and insert them into the original video after reversing the order,thus changing the video semantics.Inspired by the binary coding order in the image local binary pattern,a local binary pattern coding method about the video direction is designed to obtain a series of forward and backward coded frames.Then we match the encoding frames in both directions for correlation.And the number of matches is used to determine the authenticity of the video and locate the location of the forgery.The experimental results show that the algorithm outperforms existing methods with high robustness in the presence of significant blurring and illumination.