Video Intra-frame Forgery Detection Based On Moving Targets

With the rapid development of Internet technology and mobile devices, digital video is becoming indispensable in people’s life and the amount of data has achieved the status of explosive growth. Now, the digital video attracts more and more attention as judicial evidence. However, with the development of multimedia processing technology, such as video editing software, it has become easier to tamper the digital video and images, which resulting the multimedia information unreliable. So it is difficult to guarantee the authenticity of the video if we use the tampering video as judicial evidence. Therefore, digital video forensics has become a very important research issue and can be classified into two different categories: inter-frame forgery detection and intra-frame forgery detection. For the former, more and more scholars and engineers to enter this field and have achieved good scientific research. As inter-frame forgery, instead, is a new research direction and needs further exploration.In fact, we human is more sensitive to the moving object in the videos, thus it is a common phenomenon to tampering the moving objects. For example, copy the moving object in consecutive frames to another continuous intra-frame. In view of this, a novel technique to detect moving targets copy-move forgery in video sequences based on LK optical flow and the pre-screening mechanism of moving targets was proposed.First, the introduction of the pre-screening mechanism to detect and track the moving objects is proposed. For all the moving objects in the video which appeared in chronological order, we will use the background modeling algorithm and Kalman filter to track them and record the active region. This mechanism can filter a large number of real video frames, which can significantly improve the performance of the algorithm and reduce the time complexity.Second, for the existence of the tampering mode that is copying the moving object in consecutive frames to another continuous intra-frame, thus the two motion sequences with high correlation. To screen duplicated candidates in the temporal domain, the optical flow features in each motion sequences were obtained using the Lucas_Kanade algorithm and calculate their correlation; To evaluate the similarity of image content, the scale invariant feature transform(SIFT) algorithm is used to measure spatial correlation of each corresponding frame between the motion sequences.