Multiple forgery detection and localization technique for digital video using PCT and NBAP

Abstract

With the use of easily available video editing gadgets and applications, the authenticity of digital video has now become a significant concern since the last decade. Digital video forgeries based on inter and intra-frames are often used these days for generating false/unlawful means of evidence in many civil/criminal trials. In this paper, a passive technique using the Polar Cosine Transform (PCT) and Neighborhood Binary Angular Pattern (NBAP) along with the GoogleNet model is proposed to detect and localize multiple forgeries in the video. The pre-processing algorithm is used to extract the frames from the input videos and transform 3-D RGB color space to 2-D grayscale space. The features are then extracted using the PCT and NBAP methodologies. The extracted features are then fed to the pre-trained GoogleNet model to identify the forgeries in the video. The proposed technique can detect and localize both intra-frame forgeries and inter-frame forgeries present in the video. Besides, the proposed technique is robust enough to handle the multiple forgeries in the video, even if the video is subjected to noise exposures. With the help of performance measures like accuracy, specificity, precision, recall, and F1-score, we conduct a comprehensive performance assessment to validate the effectiveness of the proposed technique.