AttentiveFP: An Attention-Guided Deep Learning Approach for Fingerprint Liveness Detection

Biometric security systems face an increasing threat from artificial fingerprint spoofing. This research explores the effectiveness of attention-based techniques in differentiating genuine fingerprints from counterfeit ones. We introduce AttentiveFP, a deep learning approach that utilizes a dual-stream framework to analyze both holistic fingerprint patterns and localized discriminative details. By incorporating spatial attention mapping and targeted feature refinement, our method enhances the detection of subtle differences between real and fake fingerprints. Evaluations of the LivDet-2015 dataset demonstrate that our approach achieves outstanding accuracy with Cross-Match sensors (ACER of 1.77%), while maintaining robust performance across various sensor types. The model is designed to optimize detection accuracy while managing computational efficiency through adaptive thresholding, making it well suited for real-world applications. Our experimental findings confirm that Attentive-FP effectively distinguishes live fingerprints from artificial imitations while ensuring efficient processing, addressing critical concerns in practical biometric authentication systems.