Fingerprint Presentation Attack Detector Using Global-Local Model

TL;DR

This paper tackles the challenge of fingerprint presentation attack detection with limited generalization to unseen materials and sensors. It introduces RTK-PAD, a global-local PAD framework equipped with a rethinking module that localizes discriminative regions by fusing a global spoofness score and locally learned cues from patches, guided by CAM-based localization. Two pretext tasks—Cut-out for global non-local feature learning and texture in-painting for local feature learning—encourage complementary representations, enabling robust cross-material and cross-sensor performance. Empirical results on LivDet 2017 show substantial improvements over state-of-the-art methods, with an ACE of 2.28% and a TDR@FDR=1% of 91.19%, demonstrating strong generalization and practical potential for real-world deployment.

Abstract

The vulnerability of automated fingerprint recognition systems (AFRSs) to presentation attacks (PAs) promotes the vigorous development of PA detection (PAD) technology. However, PAD methods have been limited by information loss and poor generalization ability, resulting in new PA materials and fingerprint sensors. This paper thus proposes a global-local model-based PAD (RTK-PAD) method to overcome those limitations to some extent. The proposed method consists of three modules, called: 1) the global module; 2) the local module; and 3) the rethinking module. By adopting the cut-out-based global module, a global spoofness score predicted from nonlocal features of the entire fingerprint images can be achieved. While by using the texture in-painting-based local module, a local spoofness score predicted from fingerprint patches is obtained. The two modules are not independent but connected through our proposed rethinking module by localizing two discriminative patches for the local module based on the global spoofness score. Finally, the fusion spoofness score by averaging the global and local spoofness scores is used for PAD. Our experimental results evaluated on LivDet 2017 show that the proposed RTK-PAD can achieve an average classification error (ACE) of 2.28% and a true detection rate (TDR) of 91.19% when the false detection rate (FDR) equals 1.0%, which significantly outperformed the state-of-the-art methods by $\sim$10% in terms of TDR (91.19% versus 80.74%).

Figures (6)

• Figure 1: Example of inputs for the fingerprint PAD methods using entire images or patches. The first row shows images from LivDet 2017 mura2018livdet. The second row shows the resized images as input for the method nogueira2016fingerprint. The third row shows the patches centered and aligned using fingerprint minutiae, which are the inputs of the patch-based method chugh2018fingerprint. Columns (left to right): fingerprint images derived from sensors called Orcathus, GreenBit, and DigitalPersona respectively.
• Figure 2: Samples of rethinking strategy: (a). Observation of rethinking strategy used in the vision system of human beings cao2015look. (b). Our proposed scheme of rethinking strategy for fingerprint PAD. Person 1 in (a) corresponds to the local PAD module in (b), while Person 2 refers to the global PAD module.
• Figure 3: Flowchart of the proposed method, denoted as RTK-PAD. RTK-PAD consists of three modules: (a). Global PAD module, (b) Rethinking module, and (c) Local PAD module.
• Figure 4: The flow chart of the rethinking module. While, the input of this module is different scale feature maps and the prediction of global PAD module, the output is L-CAM and S-CAM.
• Figure 5: The in-painting samples, including inputs, in-painting images and ground truth. The first row shows the samples obtained from GreenBit. The second row refers to the samples derived from DigitalPersona. And the third row represents the samples captured by Orcanthus. In each block, the left image is the input, the middle one refers to the in-painting result and right one is the ground truth.