Research On Passive Forensics Of Digital Video Intra-Frame Object Tampering

With the advancement of science and technology,powerful video editing software becomes popular,people can edit and modify digital video very conveniently.To distort facts,invert black and white and undermine social order,the social information security will be seriously damaged if the content of digital video is tampered.For this reason,it is necessary to research the digital video forensics and it has the very important practical significance.In fact,digital video forensics technology could be break down into two categories,that is,active forensics and passive forensics.The digital video active forensics technology requires that the human non-perceptible information shouled be pre-embedded in the digital video.Then it determines whether or not the digital video is tampered by detecting the integrity of the embedded information.Currently,a large number of digital videos are generated every day.Many digital videos do not have the conditions to embed the information that cannot be perceived by humans.Therefore,digital video active forensics technology has significant limitations.In addition,digital video passive forensics technology is also known as the digital video blind forensics technology.In such technology,it determines whether or not the video has been tampered by analyzing the tampered traces of forensics video or detecting the changes in the statistical characteristics of the original video data.Because of it does not depend on the embedded information,the digital video passive forensics technology completely analyzed according to the characteristics of the digital video data.Therefore,it has a broad applications and has the important research value.Thus,this thesis focuses on digital video passive forensics technology.Digital video can be classified into inter-frame tampering and intra-frame tampering according to the different methods of tampering.Digital video intra-frame tampering is also called spatial tampering.In order to change the original content of the video,the object in the video frame will be modified or the different videos are synthesized in one frame.Due to its features,for example,big data and high feature dimensions,how to reduce the amount of data and feature dimensions in digital videos and improve detection efficiency is extremely important.This thesis proposes two tamper detection algorithms for intra-frame object removed and intra-frame copy-move tamper operations on digital video spatial.The main works include the following two aspects:(1)A tamper detection algorithm based on PCA(Principal Component Analysis)is proposed for object deletion and manipulation in digital video frames.To reduce the influence of noise on feature extraction of subsequent differential frames,the algorithm uses sparse representation to denoise the measured frames.Then,in order to obtain the eigenvector space of the difference frame pixels,the PCA is used to reduce the dimensions of the non-overlapping divided frames and extract the features of the pixels of the difference frames.Then the eigenvector space is classified by using the kmeans clustering algorithm,and the classification results are stored in a binary matrix.Finally,the open operation is used to operate the binary matrix to obtain the detection result.The proposed algorithm can denoise and reduce the dimension of video frames to improve the video detection efficiency.The object processed by the algorithm is the gray value of the pixels.It can make the detection effect of the algorithm not affected by the change of the video quality,and improve the robustness of the algorithm.Additionally,the experimental results show that the precision and recall are 91%and 100%respectively,and the F1 value is 95.3%,which are better than those the video forgery detection algorithm based on compression perception to some extent.Moreover,the proposed algorithm has good robustness to lossy compressed video.(2)To detect the copy-move tamper operation in digital video frames,a tamper detection algorithm based on Hessian Matrix is proposed.Firstly,the algorithm calculates the Hessian Matrix of the video frame to obtain the Hessian Matrix eigenvalue of the video frame.Then in order to remove the secondary information of the video frame,the video frame is filtered by using a filter which is based on the Hessian Matrix eigenvalue.Then SIFT(Scale Invariant Feature Transform)is used to obtain the key points(feature points)of the filtered video frame.In order to detect whether the video frame has been tampered,a neighborhood search method is used to match the key points and count the number of matched key point pairs.Finally,the tampered region is located by using the kmeans algorithm.The proposed algorithm can reduce the number of key points extracted by using the SIFT,improve the operating efficiency of the algorithm and maintain good detection accuracy.By using the proposed algorithm,the experimental results show that the detection accuracy is 92.9%,which is higher than those the HOG-based video forgery detection algorithm.