Biometric Face Presentation Attack Detection with Multi-Channel Convolutional Neural Network (1909.08848v1)

Abstract: Face recognition is a mainstream biometric authentication method. However, vulnerability to presentation attacks (a.k.a spoofing) limits its usability in unsupervised applications. Even though there are many methods available for tackling presentation attacks (PA), most of them fail to detect sophisticated attacks such as silicone masks. As the quality of presentation attack instruments improves over time, achieving reliable PA detection with visual spectra alone remains very challenging. We argue that analysis in multiple channels might help to address this issue. In this context, we propose a multi-channel Convolutional Neural Network based approach for presentation attack detection (PAD). We also introduce the new Wide Multi-Channel presentation Attack (WMCA) database for face PAD which contains a wide variety of 2D and 3D presentation attacks for both impersonation and obfuscation attacks. Data from different channels such as color, depth, near-infrared and thermal are available to advance the research in face PAD. The proposed method was compared with feature-based approaches and found to outperform the baselines achieving an ACER of 0.3% on the introduced dataset. The database and the software to reproduce the results are made available publicly.

• The paper introduces a multi-channel CNN framework that fuses sensor modalities to significantly reduce ACER to 0.3% on the WMCA dataset.
• It leverages complementary data from color, depth, NIR, and thermal channels to robustly detect sophisticated presentation attacks such as silicone masks.
• The study offers public access to the WMCA dataset and source code, supporting further research and benchmarking of PAD systems.

Biometric Face Presentation Attack Detection with Multi-Channel Convolutional Neural Networks

The paper presents a novel approach to enhance the robustness and accuracy of biometric face presentation attack detection (PAD) using a Multi-Channel Convolutional Neural Network (MC-CNN). This approach addresses the growing concern of face recognition systems being susceptible to sophisticated presentation attacks (PAs), such as those involving silicone masks, which challenge the system's ability to discern fake from genuine. The authors propose leveraging a multi-channel approach, incorporating data from various sensors, to improve the reliability of PAD systems.

Key Contributions and Findings

The main contributions and findings of this research can be summarized as follows:

1. Multi-Channel Approach: The paper introduces a multi-channel CNN framework that exploits various imaging modalities including color, depth, near-infrared (NIR), and thermal data to enhance the detection of presentation attacks. Each channel provides complementary information that, when fused, improves the system's ability to detect attacks over a single-channel approach.
2. WMCA Dataset: The research introduces the Wide Multi-Channel Presentation Attack (WMCA) database, which includes a variety of 2D and 3D presentation attacks. The dataset consists of complex and challenging attacks, such as rigid and flexible masks, providing a robust testbed for benchmarking PAD systems.
3. Detection Performance: The proposed MC-CNN achieved an impressive Average Classification Error Rate (ACER) of 0.3% on the WMCA dataset, significantly outperforming traditional feature-based methods. The results suggest that the high accuracy stems from making use of the joint representation obtained from multi-channel data rather than relying on visual spectrum data alone.
4. Capability Against Unseen Attacks: The framework was evaluated on unseen attack scenarios by leaving out one type of attack during training. This shows the method's potential to generalize to new and previously unencountered PAs, which is crucial for real-world applications.
5. Public Availability: Both the WMCA dataset and the source code for the proposed method are made available to the public, which supports the broader research community in validating and extending this work.

Implications and Future Directions

The implications of using a multi-channel approach in face PAD are noteworthy. As biometric systems move towards unsupervised applications, the ability to reliably detect and mitigate presentation attacks becomes critical. This work demonstrates that combining different imaging modalities can provide the system with a more robust feature set, effectively making it difficult for attackers to create PAIs that can simultaneously fool all channels.

Looking forward, further research might focus on optimizing the selection of sensor modalities for specific deployment scenarios, considering cost and practicality constraints. There is also potential to explore advanced integration strategies for the multi-channel data, possibly incorporating domain adaptation techniques to handle variation across different sensors. Lastly, expanding the framework to other biometric modalities could be a promising avenue to explore, with the possibility of developing a more universal anti-spoofing detector.

By addressing the challenges associated with single-channel PAD systems, this research contributes significant advancements towards enhancing the security and trustworthiness of biometric authentication systems in the face of increasingly sophisticated attacks.