Research On Passive Forensics Algorithm Of Digital Video Forgery Detection

Nowadays digital video is widely used in social fields and becomes a powerful information transmission medium.With the development of computer technology,various video editing software quickly spread and people can purposely tamper with videos without any professional knowledge.Once the forged videos are presented as court evidence,it will lead to serious consequences,undermine judicial justice and affect social stability.In order to identify the authenticity and integrity of digital video contents,video forensics has become one of the most important research topics in the field of multimedia information security attracting more and more researchers.The research of video forgery detection is the earliest and the practical application value is also the most significant.At present,many proposed algorithms need to extract respective features of original frames and tampered frames for comparison detection,that is,we must obtain both the original video and tampered video,so being difficulty to be practice applied.Based on the mature digital image forensics methods and video unique features,this thesis focus on the direct blind detection of suspected video and realizes passive forensics of intra-frame forgery and inter-frame forgery by extracting abnormal features.Intra-frame forgery is manipulated within a particular frame.As for intra-frame copy-paste forgery,this thesis proposes a passive forensics method based on edge extraction and feature points match.Because of the high correlation between the original area and the pasted one,after preprocessing of the sampled frames,this thesis uses edge detection operators to analyze the same edge lines.Meanwhile,this thesis adopts SIFT algorithm to detect and extract feature points in the sampled frame.Based on cosine similarity measure,a new feature vector matching method for feature points clustering is presented.Compared with 2NN method,our proposed method has obvious advantages in speed performance and the detection accuracy of our method is higher.Experimental results demonstrate that the proposed method can effectively detect same area blocks in the sampled frames,identify the shape and size of the copied areas and accurately locate the position of cloned region blocks.Video inter-frame forgery including frame insertion,deletion and duplication will change the original positions of frames.This thesis proposes a passive forensics method for video inter-frame forgery based on time domain similarity analysis.Adopting HSV color histogram as video frame similarity measure,we respectively calculate H-S two-dimensional histogram and S-V two-dimensional histogram of every frame and compare the histogram distance of adjacent frames.According to the anomalous changes of histogram distance,this method can accurately detect inter-frame insertion,deletion and duplication forgery.Based on the forged position,we further realize forensics double-checking of tamper types using feature similarity matching.Moving objects in video often attract people's key attention.This thesis also proposes a passive-blind forensics scheme to detect inter-frame forgery based on kinematic continuity analysis.The kinematic behavior of video object is determined by the real motion,but it also will be altered by the tampering behavior.This thesis adopts Gaussian mixture model and ViBe algorithm to detect and extract video moving objects.We detect and analyze frame deletion forgery by extracting the abnormal trajectory of the moving target.Based on four neighborhood search algorithm,we calculate centroid coordinates of moving object region and successfully detect frame duplication forgery by analyzing the abnormal change of centroid coordinates parameters.Meanwhile,according to the abnormal change position of motion trajectory and centroid parameters,the method can calculate the corresponding tampering locations.