Video copy-move forgery detection scheme based on displacement paths

Sophisticated digital video editing tools has made it easier to tamper real videos and create perceptually indistinguishable fake ones. Even worse, some post-processing effects, which include object insertion and deletion in order to mimic or hide a specific event in the video frames, are also prevalent. Many attempts have been made to detect such as video copy-move forgery to date; however, the accuracy rates are still inadequate and rooms for improvement are wide-open and its effectiveness is confined to the detection of frame tampering and not localization of the tampered regions. Thus, a new detection scheme was developed to detect forgery and improve accuracy. The scheme involves seven main steps. First, it converts the red, green and blue (RGB) video into greyscale frames and treats them as images. Second, it partitions each frame into non-overlapping blocks of sized 8x8 pixels each. Third, for each two successive frames (S2F), it tracks every block’s duplicate using the proposed two-tier detection technique involving Diamond search and Slantlet transform to locate the duplicated blocks. Fourth, for each pair of the duplicated blocks of the S2F, it calculates a displacement using optical flow concept. Fifth, based on the displacement values and empirically calculated threshold, the scheme detects existence of any deleted objects found in the frames. Once completed, it then extracts the moving object using the same threshold-based approach. Sixth, a frame-by-frame displacement tracking is performed to trace the object movement and find a displacement path of the moving object. The process is repeated for another group of frames to find the next displacement path of the second moving object until all the frames are exhausted. Finally, the displacement paths are compared between each other using Dynamic Time Warping (DTW) matching algorithm to detect the cloning object. If any pair of the displacement paths are perfectly matched then a clone is found. To validate the process, a series of experiments based on datasets from Surrey University Library for Forensic Analysis (SULFA) and Video Tampering Dataset (VTD) were performed to gauge the performance of the proposed scheme. The experimental results of the detection scheme were very encouraging with an accuracy rate of 96.86%, which markedly outperformed the state-of-the-art methods by as much as 3.14%