Research On Digital Video Inter-Frame Forgery Passivedetection Algorithm Based On Visual Content

With the fast development of the Internet and digital equipment,digital videos are widely integrating into our social life. As a newinformation carrier, digital videos enrich people’s life greatly; meanwhileit has become a significant resource in court authentication. However,with the continuous progress and innovation of video edit software, it hasbecome easily to tamper a video without leaving visual tracks, so moreand more videos are being tampered and spreading among the network.The integrity and authenticity of digital videos will influence the fairnessof the low, so how to identify if a given video is the original one withoutany forgery is of great importance in the information security field.In the paper, two algorithms to detect digital video inter-frameforgery have been proposed. One is based on digital video shot boundarydetection; the other is based on the video optical flow consistency.Video shot boundary detection based frame insertion detectionalgorithm: in surviallance videos, frame insertion will cause theinconsistency of visual content, and it will induce two video shotboundaries, which looks like a video containing several shot boundaries.By detecting the shot boundaries, the video will be divided into severalshots, and if frames in one of the shots is different than that before andafter it, then it means that frames in this shot is inserted. The video shotsegmentation algorithm contains two steps. In the first step unevenblocked color histogram difference and uneven blocked pixel valuedifference are extracted and applied to adaptive binary searching methodto detect possible video shot boundaries. After the first step, the ScaleInvariant Feature Transform is proposed to re-detect these possible boundaries returned from the first step so as to exclude fault detections.This paper proposed a frame insertion detection algorithm by firstdetecting video shot boundaries, then comparing the difference amongdifferent video shots, and locating the frame insertion points. Experimentsshow that this algorithm can achieve good performance in detection frameinsertion video forgery.Optical flow consistency based video forgery detection algorithm:this algorithm is based on the assumption that optical flows in an originalvideo are consistent in both spatial and temporal domain. However, frameinsertion, deletion and duplication will disrupt this optical flowconsistency. By detecting if the optical flows in a given video areconsistent or not, frame insertion, deletion and duplication can be welldetected. What’s more, different tampering models will affect the opticalflow differently: frame deletion will cause only one optical flowinconsistent point while frame insertion will cause two optical flowinconsistent points. There are still some other features which can helpidentify the forgery model and locating forgery points. To evaluate theefficiency of this digital video forgery algorithm, various tampered videosbased on three video databases are built manually so as not to leavevisible marks. Experiments show that this algorithm achieves greatperformance in identifying forgery and locating tampering positions.