The Forensics Of Video Processing History Based On Forensic Hash

The digital nature of multimedia data, along with powerful image and videoediting tools, has made the content of multimedia data easy to be changed, withoutleaving any obvious trace of tampering. Conventional perceptual hash and passiveforensics are two valuable tools for evaluating data trustworthiness, but they havelimitations either in terms of the scope of forensic questions that can be answered orthe computational complexity. In addition, video forgery usually compromises severalprocessing operations and steps. Many traces left by video operations may beremoved or altered by further post-processing. In response to this, researchers haveproposed an image forensics based on forensic hash. Forensic Hash is a shortsignature that is attached with multimedia, and it uses as side information to assistmultimedia forensic analysis and trustworthiness evaluation and reveal imagecomplete processing history.In this thesis, we start from researching on the forensic hash. We extract theoriginal digital content as a prior knowledge, especially when the side information isextracted. Forensics hash components are constructed, and then apply it to videoforensics for accurately revealing and recovering more information about processedhistory in terms of the likely types and parameters of the processing operations. Themain works are summarized as follows:Firstly, to estimate the processing history, we propose a video forensic schemebased on forensic hash. It analyzes the artifacts of common video tampering, andimage forensic hash is extanded to video forensics.T he statistical features of severaltypes of video tampering are extracted as side information. Forensics hashcomponents are constructed and a scalable framework of forensic hash is built in amodular way. The framework obtains robustness against content-preservingmanipulations, and an algorithm of seam-carving for video retargeting can be used tointerfere with the progress of the forensics. Experimental results show that theproposed approach can achieve the processing estimation of many tamperingoperations. Moreover, it can realize rough tampering localization with a relatively lowcomputation complexity.Secondly, a method of video seam-carving estimation based on forensic hash isproposed. Seam-carving based content-aware retargeting has been used to intentionally alter image or video content for deleting or modifying object. Themethod extracts feature points from every spatiotemporal images in the Y×T planeand each feature point set of spatiotemporal image represents a matching surface. Todemonstrate the performance of our method for all video seam-carving retargetingalgorithms, we locate the seams that represent the object removed from the frame bycomparing the distance between every adjacent matching surface pairs. Experimentsshow that the proposed hash can accurately estimate the amount of seam-carving andtheir approximate locations. Moreover, the method is not sensitive to dropping framesand detects location of adding or deleting frames by increasing the hash length.At present, the multimedia processing history research is still in its prelimi naryexploration phase. We hope that our work can promote the further development offorensics technology about video processing history.