AI-Powered Deepfake Detection and Liveness Detection

Abstract: Advances in deep learning have made it easier to create realistic visual content, popularly known as deepfakes. While such tools find applications in creative media, the serious risks of identity theft, misinformation, and digital manipulation also come with them. This work presents a lightweight detection framework that identifies manipulated visual content and verifies whether the source is a real live person. The system proposed here is empowered with both MobileNet and custom CNN models to analyze facial behavior, expression dynamics, and minute texture variations that distinguish genuine recordings from spoofing attempts created using printed images, masks, or replayed clips. For real- time processing of both images and videos, a web-based interface is developed using Flask. Experimental evaluations demonstrate accuracy close to 90%, thus extending the applicability of the proposed solution to secure authentication environments and digital forensics.

Keywords: Deepfake Detection, Liveness Detection, MobileNet, CNN, AI, Flask Web Application.