| Title |
Difference Cross-Fusion for Multi-modal Face Anti-spoofing |
| Authors |
김한별(Han-Byul Kim) ; 김종옥(Jong-Ok Kim) |
| DOI |
https://doi.org/10.5573/ieie.2026.63.1.3 |
| Keywords |
Face anti-spoofing; Difference Cross-Fusion; Multi-modal; Attention; Cross-attention |
| Abstract |
Face Anti-Spoofing(FAS) is essential for maintaining the integrity of face recognition systems in security-critical applications. In recent years, multi-modal (i.e., RGB, infrared, and depth) FAS approaches have gained much attention due to their enhanced ability in capturing diverse spoofing cues. However, existing methods often rely on simple cross-attention or extracting common features between different modals, which leads to limitations in fully exploiting the detailed characteristics of each modal. In this paper, we propose Difference Cross-Fusion, which is a simple yet highly effective module that achieves robust performance compared to existing attention methods. Our approach enables the interaction between more informative representations of each modal by transferring exclusive and modality-specific information. This plug-and-play module offers a practical and efficient solution to the growing challenges in FAS, providing a promising direction for future research and real-world deployment. |