A Novel Framework based on a Hybrid Vision Transformer and Deep Neural Network for Deepfake Detection

Generative Adversarial Networks (GANs) have enabled the creation of photo-realistic images from random noise. GAN based technologies however, led to the dissemination of synthetic images, often containing inappropriate and miss leading content, on social media. Detecting such manipulated images is crucial, yet challenging. The issue is compounded by the fact that GAN-generated images can be indistinguishable from authentic ones, rendering traditional forgery detection techniques ineffective. Deepfake images further exacerbate this problem, posing threats to news integrity, legal proceedings, and societal security. To address these challenges, we harness the potential of Vision Transformer (ViT) in conjunction with Convolutional Autoencoders (CAE) to craft innovative Framework for image analysis and deepfake detection. We introduce two distinct models, each offering unique insights into image processing. The proposed models yield excellent accuracy rate of approximately 87%, reaffirming the robustness and consistency of the proposed approach and enhanced performance compared to state of the art.