Slow Video Detection Based on Spatial-Temporal Feature Representation

As the carrier of information, digital video plays an important role in daily life. With the development of video editing tools, the authenticity of video is facing enormous challenges. As an inter-frame forgery, video speed manipulation may lead to the complete change of the video semantics. In this paper, in order to achieve effective detection for both frame sampling and frame mixing in video slow speed forgery, we proposed a spatial-temporal feature for classification. First, the periodic traces of frame difference are extracted through autocorrelation analysis, and the corresponding coefficients are used as the temporal feature. Secondly, aiming at making full use of the artifacts left in the spatial domain, and overcoming the issue of the temporal feature when the periodic traces are weak, we employ the Markov feature of the frame difference to reveal spatial traces of the forgery and utilize minimum fusion strategy to obtain the video-level spatial feature. Finally, a specific joint spatial-temporal feature is used to detect the slow speed videos through Ensemble classifier. A large number of experiments have proved the superiority of our proposed feature compared with the state-of-the-art method under two kinds of slow speed forgeries.