Trans-DF: A Transfer Learning-based end-to-end Deepfake Detector

With the advent of information and communication technologies, there have been breakthrough developments in the field of Artificial Intelligence (AI). Moreover, increasing computation power and decreasing processing times, new applications are being developed at great speeds. One such application is Deepfakes, which tackles the increased manipulated and forged media content. But these fake images and videos hamper the security and privacy of individuals and can have large-scale religious, communal, or political implications that may prove to be catastrophic for a nation. The face swapped content at times can be identified by human observation, but with the use of Generative adversarial networks (GANs), such forged content can be developed with is hard to be identified even by humans. Hence, detecting such videos and images is a challenging task for researchers. Motivated from these gaps, in this paper, we propose a pipeline for detecting and extracting human faces from videos, process them to extract features from them, and then classify them as real or fake. The results of the proposed model achieved an accuracy of 90.2% for classifying fake images from real ones.