AN ANALYSIS OF CONVNETS ON DEEPFAKE DETECTION

Abstract

The exponential growth of technology has led to massive increase in the use of images and videos as the main medium of sharing information, flooding most of the internet. However, this growth has also brought about various set of challenges, which include an increase in crimes such as identity theft, privacy invasion, and the spread of fake news through manipulated media. Generative adversarial networks (GANs) and certain kinds of data augmentation techniques are utilized to construct a face dataset comprising both real and fake examples for training the classification model. The proposed approach is a highly versatile model with very low inference and training costs, leveraging transfer learning with MobileNet and Net architectures. Evaluation is performed on two popular datasets: 140k Real and Fake Faces, and Real and Fake Face Detection, both used as popular benchmarks. The model achieves accuracy exceeding 99.5% and 75% on them. This show the e↵ectiveness of this CNN-based approach combined for detectingdigitally altered images.