The rapid advancement of face recognition technology has underscored the importance of face anti-spoofing techniques to ensure the security of these systems. While numerous deep learning-based face anti-spoofing methods have been proposed, including single classifier approaches using RGB or depth data, those based on multimodal data (RGB, depth, and infrared) have demonstrated superior performance. However, the fusion strategy for multimodal information at the score level remains underexplored. To address this gap, we present a comprehensive workflow for face anti-spoofing detection and introduce an adaptive fusion strategy for RGB and depth scores considering image quality. This approach effectively mitigates the limitations of individual RGB and depth models, enhancing their robustness against various attack types in diverse environments. We validate our method’s efficacy through tests on the CASIA-SURF, and 3DMAD datasets, comparing it with other methods. Furthermore, we demonstrate the system’s performance in real-world scenarios by testing in a realistic environment.