Adaptive Multi-Layer Facial Authentication System for Deepfake and Hybrid Spoofing Resistance: Designing A Challenge-Response System Integrating Anti-Spoofing, Liveness, and Gesture Recognition.
No Thumbnail Available
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Saudi Digital Library
Abstract
Modern authentication systems face increasing threats from sophisticated attacks including deepfakes, photo/video replay, and digital display spoofing. This dissertation develops a robust, multi-layered facial recognition system to distinguish between genuine users and sophisticated spoofing attempts. It integrates traditional computer vision techniques with advanced detection methods, including OpenPose gesture recognition, to create a comprehensive authentication solution. A formal analysis of the authentication problem demonstrates the complexity of multi-layer verification and the need for adaptive challenge-based systems. The motivation for layered security architecture and development challenges are discussed. A prototype system meeting all requirements is developed and evaluated, showing significant security improvements over traditional single-factor authentication.
Description
This dissertation presents the design and development of an adaptive multi-layer facial authentication system that defends against advanced attacks such as deepfakes, photo/video replay, digital display spoofing, and hybrid spoofing. The system integrates several computer-vision–based verification layers—including facial recognition, anti-spoofing analysis, liveness detection, gesture-based challenge-response, and deepfake detection—into a unified security framework.
Built using Java, OpenCV, and OpenPose, the prototype uses a nine-layer architecture that evaluates texture patterns, motion consistency, frequency artifacts, facial authenticity, and real-time user gestures. Evaluation results show that while single-layer facial recognition fails completely against spoofing (100% false acceptance), the multi-layer challenge-response system achieves 100% detection of static attacks such as printed photos, mobile device displays, and replay videos.
However, testing also identifies a critical vulnerability: hybrid attacks, where an attacker presents a spoofed face (photo/tablet) while performing valid gestures with their own hands. Because the system relies on a 2D camera with no depth sensing, it cannot determine whether the face and gestures come from the same physical person, leading to successful false authentications with >80% confidence.
Despite this limitation, the project demonstrates that active, multi-layer, adaptive authentication significantly enhances security and outperforms many commercial systems that rely only on passive liveness checks. The dissertation concludes with recommendations for depth-sensing hardware, improved fusion of verification layers, expanded datasets, and ethical considerations for future biometric systems.
Keywords
Adaptive multi-layer biometric authentication, biometric authentication systems, Deepfake detection, Spoofing detection, Deepfake and spoofing attacks, Gesture recognition, Facial anti-spoofing, Facial biometric system, Deepfake-resistant authentication, Active liveness verification, Gesture-based challenge–response, Hybrid spoofing detection, Real-time facial recognition, Presentation attack detection (PAD), OpenPose gesture analysis, OpenCV-based verification, Threat-resilient identity authentication, Security robustness evaluation, Anti-spoofing texture and motion analysis, CNN-based deepfake analysis
Citation
ACM, IEEE