Detection of Deepfake Video using Deep Learning and MesoNet
Lian Rebello, Linnet Tuscano, Yashvi Shah, Alvin A. Solomon, Varsha Shrivastava · 2023
Fraudsters are increasingly using evidence tampering to evade criminal charges and the acquisition of personal data for identity-related offenses. Deepfake is one of the most common strategies used today for identity theft and reputation defamation. To prevent the spread of these crimes, we need a system that can tell the difference between real and deep fake videos. Deep Neural Networks will be used in our system to identify and mark films as legitimate or manipulated, as well as the altered sections, by running the video through our trained Sequence Model, which can detect any discrepancies or alterations as a sequence of frames. LSTM will be used for temporal sequence analysis, and CNN will be employed for frame feature extraction. MesoNet is a neural network built primarily to identify deep fakes, but it would also be used for other purposes. MesoNet manages the noise produced by low-quality video processing, which impedes analysis. DeepFakes jeopardizes facial recognition and internet content. This deception is risky and can be exploited to impersonate a legitimate user. Our approach will propose a temporal-aware method for automatically detecting deepfake videos.