Deep Residual Learning for Unmasking DeepFake

Recently, DeepFake has gained a lot of popularity, any multimedia output created with deep learning technology that appears realistic to viewers is referred to as "DeepFake." Despite the positive developments of DeepFake, it has been a significant contributor to threats to an individual's privacy because it allows for the indiscernible swapping of one person's face for another without that person's permission. Additionally, it is simple for hostile actors to influence public events like elections by disseminating false information and harming national security. Therefore, identifying such DeepFake is a critical yet difficult challenge. The separation of DeepFake contents from genuine ones using the human eye has always been a challenging process, but recent research has demonstrated the use of several technologies to provide positive results for the same, but with certain limits. In order to highlight the benefits and drawbacks of the various algorithms utilized for DeepFake production and detection, the paper provides a thorough analysis of the methods employed. The proposed work focuses on using Inception-Resnet V2 to detect deep fakes and packages those advantages of deep learning for this purpose. Here the frames are collected from the uploaded video and divide it into the required number of frames in order to detect deepfakes. The subject's face is then extracted from the video using python face recognition modules. The proposed model, which has been trained on a variety of frame sequences, to determine whether the video is real or a deep fake with 95% accuracy.