Development of a video tampering dataset for forensic investigation

Abstract

Forgery is an act of modifying a document, product, image or video, among other media. Video tampering detection research requires an inclusive database of video modification. This paper aims to discuss a comprehensive proposal to create a dataset composed of modified videos for forensic investigation, in order to standardize existing techniques for detecting video tampering. The primary purpose of developing and designing this new video library is for usage in video forensics, which can be consciously associated with reliable verification using dynamic and static camera recognition. To the best of the author's knowledge, there exists no similar library among the research community. Videos were sourced from YouTube and by exploring social networking sites extensively by observing posted videos and rating their feedback. The video tampering dataset (VTD) comprises a total of 33 videos, divided among three categories in video tampering: (1) copy–move, (2) splicing, and (3) swapping-frames. Compared to existing datasets, this is a higher number of tampered videos, and with longer durations. The duration of every video is 16 s, with a 1280 × 720 resolution, and a frame rate of 30 frames per second. Moreover, all videos possess the same formatting quality (720pHD.avi). Both temporal and spatial video features were considered carefully during selection of the videos, and there exists complete information related to the doctored regions in every modified video in the VTD dataset. This database has been made publically available for research on splicing, Swapping frames, and copy–move tampering, and, as such, various video tampering detection issues with ground truth. The database has been utilised by many international researchers and groups of researchers.