An Explainable Deep Learning Framework for Robust Face Anti-Spoofing via Shapley-Guided Feature Selection
International Journal of Computational and Experimental Science and Engineering · 30.12.2025 · 11 (4)
DOI: 10.22399/ijcesen.5267 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Computational and Experimental Science and Engineering, online erschienen am 30.12.2025. Verfasst von Anisha Soni und 2 weiteren Autor:innen. Die Arbeit nennt 25 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
With biometric authentication becoming everywhere, more and more systems are targeted by advanced presentation attacks that can include physical print re-enactment and synthetic speech deepfakes. To counter these cross-modal threats, we present an elucidated multipropositional anti-spoofing system which collaboratively considers facial cues as well as acoustic cues. We compare three different visual architectures MobileNetV2, ResNet50, and ViT-L16 and an audio branch that is trained on the representations of ASVspoof speech countermeasures. To meet the urgent demand of transparency of decisions made by the network as a security application, we directly embed the Explainable AI (XAI) guided by Shapley-values directly into the network. The model actively removes spurious background activations by using custom XAIPooling and XAIDropout layers, and isolates discriminative spoofing artifacts. Using the MobileNetV2 backbone, our optimized configuration has an Average Classification Error Rate (ACER) of 3.2, an AUC of 0.98, and a test accuracy of 93.02% with MobileNetV2. These results prove that a multimodal fusion with explainable feature selection can provide a strong, interpretable, and highly efficient solution that can be easily deployed to the edge.
Zitieren
APA
Soni, A., Dubey, S., & Kushwaha, A. (2025). An Explainable Deep Learning Framework for Robust Face Anti-Spoofing via Shapley-Guided Feature Selection. International Journal of Computational and Experimental Science and Engineering, 11(4). https://doi.org/10.22399/ijcesen.5267
Vancouver
Soni A, Dubey S, Kushwaha A. An Explainable Deep Learning Framework for Robust Face Anti-Spoofing via Shapley-Guided Feature Selection. International Journal of Computational and Experimental Science and Engineering. 2025;11(4). doi:10.22399/ijcesen.5267
BibTeX
@article{Soni2025,
author = {Soni, Anisha and Dubey, Sipi and Kushwaha, Ajay},
title = {An Explainable Deep Learning Framework for Robust Face Anti-Spoofing via Shapley-Guided Feature Selection},
journal = {International Journal of Computational and Experimental Science and Engineering},
year = {2025},
volume = {11},
number = {4},
doi = {10.22399/ijcesen.5267}
}
EUVIMED