Secure Vision: Integrated Anti-Spoofing and Deepfake Detection System Using Mobilenet and Resnext

Abstract

In recent times, new phenomena such as deepfake videos and facial spoofing attacks imply new problems for security systems based on biometric authentication and digital content checks. The Secure Vision project provides a single solution for both spoofing attack and deepfake detection by using a deep learning model. Real-time anti-spoofing is accomplished using MobileNet that is trained for detecting liveness clues from facial images whereas ResNeXt is used for deepfake detection due to its capability to detect artefacts and discrepancies in the manipulated videos. With these two approaches integrated, Secure Vision can offer a single efficient yet optimised solution of securing visual media content and authentication processes in real-time. The experimental results prove that both the model’s overall accuracy is considerable, whereas MobileNet reached 96.2% identification rate for spoofing and ResNeXt realised 94.5% rate of identification for deep fakes. This system has potential in privacy and access, physical and cyber security, social media, identity theft and fake news.