Deep Guard: Face Spoofing Detection using Swin Transformer and rPPG Signal

Abstract

The Deep Guard project is developed to detect and prevent facial spoofing attacks in biometric authentication systems. Traditional face recognition systems are often vulnerable to spoofing attempts using printed photos, replayed videos, or 3D masks. To overcome these challenges, Deep Guard integrates Swin Transformer-based deep feature extraction with rPPG (remote Photoplethysmography) signal analysis to accurately differentiate between real and fake faces. The Swin Transformer captures fine-grained spatial and texture information, while the rPPG module extracts heartbeat-based color variations to confirm liveness. The extracted features are then fused using an advanced feature fusion technique for robust classification. A trained model finally classifies the input as genuine or spoofed. Experimental evaluation shows that Deep Guard delivers high precision, adaptability, and real-time performance, making it a reliable and secure solution for modern facial authentication applications in banking, mobile security, and access control systems.

Introduction

Facial recognition is widely used for authentication due to its convenience and accuracy but is vulnerable to spoofing attacks using photos, videos, or 3D masks. To address this, the Deep Guard system combines visual and physiological analysis to enhance anti-spoofing reliability.

Key Components & Methodology:

1. Preprocessing: Captures facial images or video frames, detects regions of interest, normalizes lighting, reduces noise, and aligns frames to optimize visual and physiological feature extraction.

2. Visual Feature Extraction (Swin Transformer): Captures fine-grained spatial and texture details, identifying subtle differences between real and spoofed faces.

3. Physiological Feature Extraction (rPPG): Detects heartbeat-induced skin color changes to verify liveness, which is absent in spoofed inputs.

4. Feature Fusion & Classification: Combines visual and physiological features to enhance discriminative power, then classifies the face as genuine or spoofed using a neural network, providing confidence scores for authenticity.

Process Flow:

• Video/image acquisition → face detection → preprocessing → parallel feature extraction (Swin Transformer & rPPG) → feature fusion → classification → result visualization.

Results & Visualization:
Deep Guard demonstrates high accuracy, low false acceptance rates, and robustness across various spoofing types. Visualization tools display detected facial regions, confidence scores, heatmaps of key features, and pulse waveforms, enabling interpretability and real-time monitoring.

Conclusion

The Deep Guard system provides a reliable and intelligent solution for face anti-spoofing by combining Swin Transformer-based visual feature extraction with rPPG-based physiological analysis. Through the fusion of deep spatial and temporal features, it effectively distinguishes real faces from spoofed ones such as photos, videos, or 3D masks. The system demonstrates high accuracy, robustness, and real-time performance, making it suitable for secure authentication in various domains like mobile devices, banking, and access control. By integrating deep learning and liveness detection, Deep Guard significantly enhances the security and reliability of modern facial recognition systems.

References

[1] Z. Yu, X. Li, and G. Zhao, “Revisiting pixel-wise supervision for face anti-spoofing,” IEEE Transactions on Biometrics, Behavior, and Identity Science, vol. 2, no. 3, pp. 274–283, 2020. [2] A. George and S. Marcel, “Learning one class representations for face presentation attack detection using multi-channel convolutional neural networks,” IEEE Transactions on Information Forensics and Security, vol. 16, pp. 361–375, 2021. [3] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), pp. 770–778, 2016. [4] Z. Liu, Y. Lin, Y. Cao, H. Hu, Y. Wei, Z. Zhang, and S. Lin, “Swin Transformer: Hierarchical vision transformer using shifted windows,” Proc. IEEE/CVF International Conference on Computer Vision (ICCV), pp. 10012–10022, 2021. [5] W. Wang, A. den Brinker, S. Stuijk, and G. de Haan, “Algorithmic principles of remote PPG,” IEEE Transactions on Biomedical Engineering, vol. 64, no. 7, pp. 1479–1491, 2017. [6] X. Liu, J. Wan, and G. Guo, “Multi-scale CNNs for face anti-spoofing,” IEEE Transactions on Information Forensics and Security, vol. 13, no. 11, pp. 2752–2767, 2018. [7] C. Parkin and A. Gravenor, “Real-time facial liveness detection using deep learning and rPPG signals,” Pattern Recognition Letters, vol. 145, pp. 52–59, 2021. [8] T. Zhang, F. Yang, and X. Wang, “A survey on face presentation attack detection,” Neurocomputing, vol. 455, pp. 240–258, 2021. [9] S. Määttä, A. Hadid, and M. Pietikäinen, “Face spoofing detection from single images using micro-texture analysis,” Proc. IEEE Int. Joint Conf. on Biometrics (IJCB), pp. 1–7, 2011. [10] H. Li, P. Kumar, and A. C. Kot, “Face anti-spoofing with image distortion analysis,” IEEE Transactions on Information Forensics and Security, vol. 13, no. 10, pp. 2408–2423, 2018. [11] G. Heusch, A. Anjos, and S. Marcel, “Deep representations for face presentation attack detection,” Proc. IEEE Int. Conf. on Biometrics Theory, Applications and Systems (BTAS), pp. 1–8, 2020. [12] J. Hernandez-Ortega, J. Fierrez, A. Morales, and J. Galbally, “Introduction to face presentation attack detection,” Springer Handbook of Biometrics, pp. 243–262, 2022. [13] X. Yu, J. Li, and W. Deng, “Attention-based liveness detection for face recognition using RGB and temporal features,” Sensors, vol. 22, no. 4, pp. 1234–1246, 2022. [14] S. Jia and Z. Guo, “Joint spatial-temporal learning for face anti-spoofing,” IEEE Access, vol. 10, pp. 9852–9864, 2022. [15] P. Wang, Y. Chen, and H. Zhang, “Hybrid feature fusion framework for robust face anti-spoofing,” Expert Systems with Applications, vol. 210, pp. 118–126, 2023.

Copyright

Copyright © 2025 S. G. Santhoshini, K. S. Shalini, K. Usha Rani. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.