Date of Publication :29th June 2023

Abstract: Nowadays, anomalous detection is of utmost importance in videos because of unusual activity (or) unauthorized changes in the video frame. Several techniques have been involved in authenticating and localizing video tampering but most of them are inefficient. To enhance the detection of video tampering, the new Hybrid adversarial deep network (Hybrid DeepNet) has been proposed. The videos are collected from three datasets such as VTD, MFC 18 and VIRAT. This model mainly consists of five ways to identify the tampered region of videos. They are double compression detection, noise filtering process, segmentation, feature extraction and tamper detection. Initially, the double compression process is evolved using an improved group of pictures (GOP) method and then, the rapid bilateral filter is used to remove the noise involved in the video. By using wavelet-based feature extraction, the features are extracted. The new hybrid adversarial deep network with white shark optimization (WSO) has been performed to detect and localize the forged region of video. Because of the WSO algorithm with a deep model, the proposed method is helpful for removing network loss. The performance of the proposed model is evaluated by comparing with existing model. The proposed model obtained 96.21%, 95.15%, 96.13% and 97.15% for accuracy, recall, F-score and precision, respectively.

Reference :