Study On Passive Digital Video Forensics For Inter-Frame Tampering And Multi-Compression

With the vigorous development of Internet and multimedia technology,the use of digital video increases rapidly.What comes together is the development of powerful video editing tools,with which users can alter the video content easily and create visually convincing video.It becomes very important to identify the authenticity of a given video,especially when it serves as evidence in court or news item.Digital video forensics emerges and attracts more and more attention.Digital video forensics includes active forensics and passive forensics.Different from active forensics which needs pre-embedded authentication information in video,passive forensics is capable of authenticating digital videos based on the characteristics of the video itself and becoming a significant research topic in information security field.Inter-frame tampering is one of the most common forgeries in digital videos,which makes effective inter-frame forgery detection one of the key issues of passive video forensics.Furthermore,tampered videos must have been double compressed,so double compression detection becomes another key issue of passive video forensics.Therefore,inter-frame forgery detection algorithms and double compression detection algorithms have both theoretical significance and practical significance.The main innovative work of this thesis includes the following four aspects.1.Two inter-frame forgery detection algorithms are proposed to improve the detection accuracy.Firstly,based on the analysis that inter-frame tampering will decrease correlation between adjacent frames at the tampering point,quotients of correlation coefficients of LBP(Local Binary Patterns)and quotients of MSSIM(Mean of Structural Similarity)are presented to indicate the correlation change of image content.Then abnormal points can be detected by using Tchebyshev inequality and decision-thresholding,and inter-frame forgeries can thus be identified.2.In order to improve the detection accuracy of inter-frame forgeries which are difficult for visual identification,two inter-frame forgery detection algorithms are proposed.One detection algorithm is based on MVP(Motion Vector Pyramid)feature.By combining the motion vector module and image pyramid,MVP can catch the subtle changes between adjacent frames of the video,thus the variation of the two frames can be revealed more accurately.The other detection algorithm is based on post-processed MVP.With a series of post-processing operations,i.e.,mean removal,accumulation and histogram calculating,the traces left by the inter-frame tampering operation can be better captured.This can further improve the classification accuracy to distinguish frame-deleted forgeries and frame-duplicated forgeries from original videos.3.An algorithm for detecting multiple H.264/AVC compressions with the same QP,which is seldom studied in the field of video forensics,is proposed.Firstly,ratio difference set is calculated by identifying the quantized DCT coefficients whose values will be changed between consecutive compressions.Then a feature set including quartiles of ratio difference set is constructed to serve as input for classification.With the aid of SVM(Support Vector Machine)classifier,the feature set can be used to classify multiply-compressed videos from singly-compressed videos.Experimental results show that the proposed algorithm has high classification accuracy and robustness against copy-move attack and frame-deletion attack.4.An algorithm for detecting double HEVC compressions with different bitrates is proposed.Firstly,two feature sets of I frames are analyzed for singly-compressed and doubly-compressed HEVC videos.One is the number of PU(Predcition Unit)blocks feature set based on partition types of PU,and the other one is first digit distribution feature set based on the quantized DCT coefficients.With combination of the two feature sets,the traces of double compression can be better characterized.Experimental results show that the proposed algorithm can effectively distinguish doubly-compressed videos from singly-compressed ones.Based on the mutation characteristics of image contents at the tampering point,quotients of correlation coefficients of LBP and quotients of MS SIM which reflect image content are proposed and MVP which reflects subtle change between adjacent frames is further proposed.Then by using Tchebyshev inequality and Generalized ESD test,inter-frame forgery detection algorithms which distinguish inter-frame modified videos from original ones have been studied in this thesis.In addition,based on the lossy characteristics of compression coding,double compression detection algorithms have been discussed from H.264/AVC coding standard to the latest HEVC coding standard,which would provide a variety of ideas for the research of video double compression detection.