Forensics Of Digital Video Based On Visual-content Congruency

With the entering into the21st Century, and the rapid development of network, broadband speeds, hardware storage technology and multimedia technology, it becomes convenient to watch and get access to online video by mobile devices. However, at the same time, many kinds of video editing software prevails, which makes it easy for non-professionals to edit video, and even target tamper digital video, while it is the privilege of professionals in the past. Once tampered video adopted by the official media, insurance, court, that will lead to people misunderstand about the facts and produce an unpredictable impact on society. As a result, detecting the authenticity and integrity of videos becomes the most concerned problem and an important study direction in the multimedia field.The existing ways of the video forgery is divided into time domain tampering, spatial domain tampering, and the joint tamper with the two ways. In view of the three ways of video forgery, the detection can be classified into two types:the active detection and passive detection. The active detection mainly detects the digital watermark and digital signature by using image processing technologies. But it requires the shooting with mosaic function inside the equipment, and it is invalid with the tampered video of no digital signature or watermark. So it becomes the focus of attention for us to find passive tamper detection algorithm for video tamper, which only relies on the characteristics of the video itself.The detection digital video forgery is newly emerging in recent years. The study is in the initial stages and there are still many shortcomings of the achievements we have made. Therefore, this article focuses on tampers of frame duplication, deletion and insertion in time domain. After the summary of three existing tamper detection technologies, two different detection algorithms for tampered video are proposed. The major works of the thesis of Master are as follows:(1) So far, there is no public and unified database of tampered video for verifying the rationality and validity of the proposed algorithms. In this paper, firstly,8original YUV format video clips are downloaded from the Internet. Then two MOV format video clips are taken by using the different types of camcorder. Thus, the test database composes of30tampered video clips by tampering on the10initial video clips.(2) This paper presents the advantages and the problems of the existing algorithms, after analysis of the current passive video tamper detection algorithms.(3) A new digital video forgery detection algorithm was proposed which was based on multi-scale normalized mutual information. The algorithm innovatively introduces the information theory of mutual information to the digital video forgery detection domain for the first time, and build a model to the measure similarity of visual content; Secondly, a content congruency model based on scale space theory was proposed; Thirdly, the abnormal value model of multi-scale normalized mutual information descriptor (MNMI) is build by the LOF algorithm; Finally, the MNMI abnormal value which is more than a certain threshold, will be judged abnormal, correspondingly, the number of the video frame will be judged tampered position. The experimental results demonstrate that the proposed algorithm is effective.(4) A digital video forgery detection algorithm based on color-content congruency was proposed. In this algorithm, the color histogram is used to describe the visual content because of the rotationally invariance and scale invariance. This feature has the advantage of the stability for any size of brightness and direction images. Secondly, the histogram intersection is used to compute the similarity of video frames. Finally, the adaptive threshold method is adapted to determine whether the video has been tampered. The experimental results demonstrate that the proposed algorithm is effective.