Video Local Tampering Detection And Video Forgery Detection Database Construction

The rapid development of computer multimedia techniques brings many opportunities to the human society as well as larger challenges to the social security.As a common network medium in daily life,the digital videos are being widely disseminated,meanwhile,they are also at risk of being modified.Some modifications are just for entertainment without malicious motivations,while people may tamper videos to achieve their sinister purposes.Thereinto,intra-frame tampering(such as facial area tampering)videos are especially difficult to manufacture and also hard to detect.Meanwhile,the rapid development of video editing software and computer technology makes intra-frame video tampering become easier.For example,the popular video tampering method known as Deepfake fabricates a face-swap video involving no more than a few clicks in just a few minutes,which has brought potential safety threat to the society.Consequently,the research on video intra-frame tampering detection technology is of great significance.Concentrating on intra-frame video tampering and deepfake video,the research works included in this paper are presented as follows:1.First author analyzed the real and face-swap videos from the perspective of video coding.Specifically,author found out the differences between the real and fake videos by some encoding parameters of the video(such as motion vectors and partitions),which can be used to provide a reasonable idea for the forgery detection of deepfake videos.Moreover,to verify the rationality of the thought,some tampering detection features in traditional video or image forensics are used by author to analyze the differences between real videos and deepfake ones.2.A deepfake video tampering detection algorithm based on stability of facial temporaldomain features is proposed in this paper.With taking the inherent biometric characteristics of organs and facial contour into account,author divided the facial landmarks into three categories according to their degree of activity,which can be used to distinguish between the real and deepfake video frames.In this paper,organ landmarks and contour landmarks are directionally connected to construct facial vectors which are first proposed by author to describe the relativeness between organs and contour positions.The inconsistency caused by the generation mechanism of deepfake videos is used to judge the authenticity of the video by calculating the deviation angle of the facial vector between two successive frames.The experiment shows that the algorithm has a good detection result on four popular databases.The algorithm makes use of the inherent characteristics of biological signals for detection,which improves the detection effect and is proved to be of good commonality.3.On account of the lack of comprehensive video forgery detection databases,we construct a video forgery detection database in which all videos are recorded or tampered by us.The database contains videos shot by various types of camera equipment under eight application situations.The effectiveness of the database was analyzed by seven common tampering detection algorithms.It was proved that each algorithm can exhibit its performance differences in this database,and the database is proved to be an effective database for video forgery detection.