Hybrid Face Spoof Detection Framework Using HAAR Feature Analysis and 3D Depth Estimation

Abstract

Face spoofing attacks pose a significant threat to biometric authentication systems, as facial recognition models can be deceived using printed photographs, digital screen displays, or replayed videos. Traditional face recognition techniques primarily focus on identity matching and often neglect liveness verification, making them vulnerable under diverse environmental conditions. To address this limitation, a hybrid face anti-spoofing framework is proposed that integrates structural feature validation, pseudo-depth estimation, and temporal consistency analysis for robust real-time liveness detection. The system employs Haar Cascade-based facial feature verification to ensure structural integrity, while gradient-based pseudo-depth estimation analyzes surface curvature variations to distinguish real three-dimensional faces from planar spoof media. To enhance stability across consecutive video frames, a temporal fusion mechanism based on the Exponential Moving Average (EMA) is implemented to smooth classification outputs and reduce prediction fluctuations. The proposed system is implemented in MATLAB and evaluated under real-time webcam conditions using both genuine and spoof samples. Experimental results demonstrate improved robustness against print and replay attacks compared to single-feature approaches, achieving an overall accuracy of 90% with reduced false acceptance rates. The lightweight architecture ensures low computational complexity, making it suitable for real-time biometric authentication and edge-based deployment.