Research On Passive Forensics For Video Intra-frame Forgery

With the rapid development of multimedia technology,various kinds of powerful video editing softwares have been widely used.With the help of these tools,people can easily edit or tamper with video content.If these tampered videos are illegally used,some negative consequences may even affect social stability.Therefore,there is an increasing research interest in video forensics,which is used to authenticate the veracity and integrity of videos.Compared with image,the properties of digital video such as huge volumes,high dimensionality and nonlinearity in data,have brought more challenges to video forensics.Although many researches on the field of video passive forensics have been done in recent years and a variety of forgery detection algorithms have been proposed,these algorithms generally suffer from low accuracy,high complexity and low robustness.In this dissertation,based on the research status above,two most common types of video forgery,i.e.,region duplication and foreground removing are studied,and new video forgery detection algorithms have been proposed,which were all simulated and compared by experiment to get better research results.Our main work and research contributions include the following three aspects:1.A novel passive forgery detection algorithm based on MI-SIFT(Mirror and Inversion Invariant Generalization for SIFT Descriptor)for video region duplication is proposed.In the algorithm,MI-SIFT is introduced and improved,realizing the detection of videos to which a mirroring operation was applied.Then,object tracking technique is employed to locate the tampered area,avoiding the overloaded calculation of extracting feature and matching frame by frame and improving the detection efficiency.The experimental results demonstrate the satisfactory performance of our algorithm for detecting videos subjected to mirror operations and its higher accuracy as well as efficiency than previous algorithms.2.A fast forgery detection algorithm based on Exponential-Fourier moments(EFMs)for detecting region duplication in videos is proposed.First,the algorithm improves the extraction method of EFMs and puts forward a new block-matching algorithm,improving the efficiency.Then,an adaptive parameter-based fast compressive tracking(AFCT)algorithm is proposed by improving the Fast Compressive Tracking(FCT)algorithm with two aspects to track the tampered regions,increasing the tracking accuracy and robustness.The experimental results show that our proposed algorithm not only detects videos subjected to mirror operations,but also achieves the detection of multiple regions forgery as well as moving target forgery effectively.Furthermore,our proposed algorithm is superior to those of the existing algorithms in the performance of detection efficiency,detection accuracy and robustness.3.A novel forgery-detection algorithm based on Energy Factor(EF)that identifies video foreground removing is proposed.Firstly,aiming to the feature of video foreground removing,a calculation method based on Energy Factor(EF)is put forward.By constructing the EF curve of the video,the algorithm identifies the video sequences that are forged.Secondly,an adaptive parameter-based visual background extractor(AVIBE)algorithm is applied,which is combined with EF,to extract the suspect traces from each frame in the sequences determined in the first stage.Finally,combine the results of the above two steps together and output the final detection results.Experiments indicate that,compared with the existing algorithms,our algorithm can greatly reduce the time cost,thus effectively overcoming the problem of low detection efficiency.Furthermore,our proposed algorithm has higher detection accuracy and better robustness against complex background,such as the interference of swaying trees,grasses and water waves.In conclusion,aiming at the forgery of video region duplication,the passive forgery detection algorithm based on MI-SIFT proposed in this dissertation not only has high accuracy and efficiency,bus also realizes the detection of videos subjected to mirror operations.Furthermore,in the forgery detection algorithm based on EFMs for detecting region duplication,the performance of detection speed,accuracy and robustness are further improved.Meanwhile,the applicability has been further expanded in other aspects such as multiple forgery and moving target forgery.Aiming at the forgery of video foreground removing,the forgery detection algorithm based on Energy Factor(EF)is proposed,which has the advantages of high detection speed,high accuracy and strong robustness.Therefore,the research results of this dissertation have the theory significance and application value.