| With the rapid development of Internet technology and the popularization of smart phones,it has brought tremendous changes to people’s lives.It can be said that the Internet and smart phones truly connect everyone with each other.This is an opportunity for online video can be developed faster ever.However,while people are enjoying the ease and pleasure of online videos,some criminals maliciously tamper with online videos,making videos with normal meanings become tampered videos that seriously mislead the audience’s subjective perceptions.Some of them are widely spread.The tampered video even can produce incorrect public opinion orientation and cause serious social impact.In order to reduce the harm caused by tampering with the video and effectively monitor the authenticity and integrity of the original video,a video-based forensic research algorithm comes into being.Video forensics is mainly used to detect video related technologies,which mainly include active forensics and passive forensics.Passive forensics technology does not need to add additional information to the video.The authenticity and completeness of the video can be judged only based on the information contained in the video itself,so the application fields of passive forensics algorithms are more extensive.Passive video forensics can be divided into two types: inter-frame forensics and intra-frame forensics: intra-frame forensics is the feature extraction of the frame content of a single frame,and whether the extracted features contain tampering information is used to determine whether the video has undergone intra-frame tampering.Inter-frame forensics is to observe and compare whether the correlation between frames is abnormal to determine whether there is inter-frame tampering in the video.This thesis conducted in-depth research on the video forensics algorithm and the inter-frame forensics algorithm,proposes two video passive forensics algorithms,and verifies the correctness and effectiveness of the two passive forensics algorithms through experimental results.On the basis of completing the research of video passive forensics algorithm,this paper implements a video-based passive forensics system,which can read local video files and decode the video to obtain a series of video frame sequences.Forensics are obtained frame by frame for the frame sequence,and the forensic results are displayed in a visual form of forensic statistical graphs.For the video forensics module,this system proposes and implements two algorithms:(1)One is an inter-frame forensics algorithm based on Local Binary Patterns(LBP).The algorithm first converts the input frame image into an LBP format image to better highlight the texture information of the frame image,and then in the frame sequence,take out the LBP images of adjacent frames and compare them to obtain the associated information sequence between adjacent frames.Finally,the adjacent associated information is processed to obtain the associated information fluctuation statistical sequence,and observing whether the fluctuation statistical sequence contains a large one,if it exists,it proves that the input one is a video that has been tampered with between frames,and the location where large fluctuations occur is the tampered location.(2)The other one is an intra-frame forensic deep learning algorithm based on bi-channel CNN.The algorithm proposes to use two CNNs to perform intra-frame forensics on input frame images.One of the CNNs uses the idea of steganalysis.Before forensics,the spatial rich model is used to extract the noise residuals of the input frame image to better extract the input frame.The tampering feature of the image,another CNN directly extracts the input frame image,which can better make CNN pay attention to the substantial changes in the content of the frame image due to the tampering.Finally,the output feature map results of the two CNNs are combined as the basis for the two classifications,so that the entire forensics algorithm takes into account the tampering characteristics and content characteristics of the input frame image.This system includes two video passive forensics algorithms,which can effectively protect the integrity and authenticity of digital video without adding additional information to the video. |
