.png){kind=icon}
.png){kind=icon}
Return
An explainable deep learning framework for fingerprint spoofing attack detection
.png){kind=icon}
PRE
AI
DOI：10.1007/s11334-026-00627-7.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Artificial intelligence is gaining ground in several application areas, including biometric authentication. However, several limitations and shortcomings of the learning algorithms are applied to make such systems intelligent. One of the most prominent limitations is the need for more rationality and understanding of the decisions made by the algorithms, which makes these systems vulnerable to intruder attacks. In the case of fingerprint spoofing, it is crucial to understand the reasoning behind the classifier's decisions. To address this limitation, a robust convolutional neural network (CNN) with Gradient-Weighted Class Activation Mapping (GradCAM) is proposed for detecting fingerprint spoofing attacks. Visual explanations are generated by applying GradCAM on the fused CNN layers rather than the last layer, thus improving the interpretability of images. Additionally, softmax classifier scores are subjected to threshold optimization to compute evaluation metrics, including accuracy, False Acceptance Rate (FAR), and False Rejection Rate (FRR), across predefined splits and multiple folds. The explanations generated are subjected to verbal explanations using another popular explainable framework, i.e. Local Interpretable Model-Agnostic Explanations (LIME). The framework is further validated by comparing it with various deep CNN frameworks, such as ResNet-18 and EfficientNet-v2, on different benchmarks of the LivDet 2013 datasets, which indicate its superiority over state-of-the-art approaches.
Keywords:
Explainable
Biometrics
Deep learning
LIME
Spoofing
Machine learning
Journal
IF:
1.1
Papers:
20
Citations:
416
View more organizations