Research On Forensics Algorithm Of Digital Video Inpainting Forgery

Digital videos are generally regarded as reliable records of events,and are widely used in transportation,medicine,and judicial fields to provide basic evidences.However,with the rapid development of digital video,video processing technology and software are constantly emerging.People can use the video editing software to remove unwanted local objects in the video easily.If the forged video is used maliciously,it will do harm to people's daily life and normal political,economic and social order.As an important branch of digital video forensics,video inpainting forensics has become a very important research topic in the fields of multimedia content analysis and information security.Based on the intensive study of digital video inpainting forensics algorithms at home and abroad,three forensic solutions are proposed.The main achievements include:(1)Based on analyzing the characteristics of the forged area in spatio-temporal domain,a video inpainting forensics algorithm based on spatio-temporal LBP coherence analysis is proposed.In order to process the dynamic background video captured by the mobile camera,this method first calculates the motion vector of the background area in each video frame to complete the preprocessing operation.Then the spatial LBP operator is used to perform correlation analysis on the image blocks at the same position of two adjacent video frames to find suspicious forged areas.Finally,the temporal LBP operator is utilized to remove the authentic regions that are falsely classified as forged to locate the final forged areas.The experimental results prove that the method can detect and locate the forged regions effectively and is robust to the quantization noise generated in the video compression.(2)Aiming at the limitation that the first forensics scheme does not make full use of the time domain information of the video,a video forensics algorithm based on spatiotemporal slicing is proposed.This method first performs a median filter operation on the video frame.Then the video volume is cut along the time axis to obtain the vertical spatiotemporal slice image,and the detection and localization of the forged area are realized by analyzing the time-domain inpainting traces of the slices image.For the video with slight movement in the background,the video volume is cut in the frame-by-frame fine-tuning manner.And the experimental results show that the algorithm can effectively detect and locate the videos tampered with video inpainting software.(3)Due to the shortcoming that the detection performance of artificially designed feature algorithm depends on the accuracy of feature selection,a two-channel convolutional neural network based video inpainting forensics algorithm is proposed.This method uses the fusion feature extracted from the RGB channel and the noise channel through the two-channel structure to detect the video inpainting forgery.The RGB channel is able to find the unnatural tampering boundary and contrast in the absolute frame difference image,and the noise channel is able to find the inconsistency of noise between the original region and the tampered.And the experimental results reveal that the proposed detection algorithm effectively improves the recognition accuracy of forged video frame,and compared with the traditional single-channel network structure,the pattern of two-channel feature fusion achieves better detection performance.