Leveraging Intermediate Features of Vision Transformer for Face Anti-Spoofing

2025-05-30

AI Summary

Face anti-spoofing systems remain vulnerable to presentation attacks such as printed photos and screen replay. To address this, we propose a live face detection method leveraging intermediate-layer features of Vision Transformers (ViTs). Departing from the dominant paradigm that relies solely on final classification logits, we systematically exploit shallow- and mid-level ViT features—characterized by both local texture sensitivity and global structural discriminability—for fine-grained spoof detection. We introduce an intermediate-feature distillation loss and a confidence-scoring mechanism for liveness estimation. Additionally, we design face-specific data augmentation strategies: semantic-preserving facial region enhancement and adaptive patch-wise masking. Evaluated on OULU-NPU and SiW benchmarks, our method achieves state-of-the-art performance against print and replay attacks, with significant improvements in AUC and HTER metrics.

Technology Category

• Computer Vision: Biometrics, Face, Gesture & Pose
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Feature Construction/Reformulation

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Security and Privacy: Data transparency and provenance
• Systems and Infrastructure for Web, Mobile and WoT: Sustainability and carbon-aware systems for Web, mobile, and WoT

Abstract

Face recognition systems are designed to be robust against changes in head pose, illumination, and blurring during image capture. If a malicious person presents a face photo of the registered user, they may bypass the authentication process illegally. Such spoofing attacks need to be detected before face recognition. In this paper, we propose a spoofing attack detection method based on Vision Transformer (ViT) to detect minute differences between live and spoofed face images. The proposed method utilizes the intermediate features of ViT, which have a good balance between local and global features that are important for spoofing attack detection, for calculating loss in training and score in inference. The proposed method also introduces two data augmentation methods: face anti-spoofing data augmentation and patch-wise data augmentation, to improve the accuracy of spoofing attack detection. We demonstrate the effectiveness of the proposed method through experiments using the OULU-NPU and SiW datasets.

Problem

• Detect spoofing attacks in face recognition systems
• Utilize Vision Transformer intermediate features for detection
• Improve accuracy with novel data augmentation methods

Innovation

• Uses Vision Transformer intermediate features
• Introduces two data augmentation methods
• Detects minute live vs spoofed differences