FALSIFYX: Guarding Truth in the Age of Synthetic Media

Abstract

The Deepfake technology has been a major threat to the integrity of digital media ever since its introduction in 2017. First uses were based on face-swapping models based on autoencoders and Generative Adversarial Networks (GANs), however, with quick development in synthetic media production, these original detection methods have been outpaced. This research puts forward a self-training deepfake detector that does not rely on the existence of large data sets. The system uses dynamic and confidence-based machine learning algorithms that are dynamically adjusted within user feedback. The model combines facial feature analysis using the computer vision method with audio-visual inconsistency detection to form a multimodal verification framework. It utilizes transfer learning and online learning to adjust to new variants of deepfakes at a given time. The results of experiments indicate high levels of performance in detecting abnormalities like eye-blinking abnormalities, micro-expressions, and audio-visual inconsistencies. Future developments are transformer-based time-based modeling and the use of deepfakes to detect text to enhance authenticity and provide reliable and safe online media.