Masked Face Deepfake Detection: A Robust and Accurate Framework for Modern Threats

Abstract

Increasing number of people concerned about Deepfake technology is a way to create synthetic media using artificial intelligence - and its ability to spread incorrect information and facilitate digital manipulation is focused on this study. The wide use of acute face masks of the Covid-19 epidemic, and luminously more sophisticated identity techniques, has made it even more difficult to identify Deepfake films. To fulfill this requirement, the Deepfake Face Mask Dataset (DFFMD) project presents a collection of customized images designed to train the detection algorithm, which is identifying the Deepfake Video, where Varna Masks are wearing. Including preparatory steps, function-based analysis, remaining connections and batch normalization, the study benefits from a new technology using an Inception-ResNet-v2 architecture. This advanced model improves more traditional approaches, including InceptionResNetV2 and VGG19, especially when the faces of individuals handle hidden conditions. The suggested model's potential as an effective countermeasure is highlighted by the experimental results, which demonstrate its amazing accuracy in recognizing deepfake films with face masks. Recognizing the dynamic nature of deepfake threats and the necessity for adaptable solutions, the paper calls for ongoing research to improve detection capabilities. Notable terminology that emphasize the use of advanced architectures in the quest for strong deepfake detection are Xception, Inception ResNetV2, VGG19, and CNN.