Search papers, labs, and topics across Lattice.
This paper introduces the Cascade Forgery Mining Network (CFM-Net) for Fingerprint Presentation Attack Detection (PAD), which addresses the challenge of varying Artifact Extraction Difficulty (AED) across different fingerprint regions. By quantifying AED using local Gabor feature certainty and employing an adaptive-depth feature extraction architecture, the CFM-Net enhances the detection of subtle artifacts in high-AED areas. Experimental results on LivDet datasets show that this approach significantly outperforms existing methods, particularly in classifying high AED fingerprints, thereby improving security in fingerprint identification systems.
High Artifact Extraction Difficulty regions can be accurately classified with a novel network that adapts to varying extraction challenges, leading to significant improvements in fingerprint PAD.
Fingerprint Presentation Attack Detection (PAD) is a critical component of fingerprint identification systems, serving as a protective measure against unauthorized access. In this paper, we observe that different regions of a fingerprint image can exhibit varying Artifact Extraction Difficulty (AED), with high-AED regions requiring more sophisticated extraction mechanisms to capture more subtle discriminative evidence. To address this issue, we propose to quantify AED using local Gabor feature certainty and partition fingerprint images into multiple regions based on their respective AED values. We then propose an AED guided Cascade Forgery Mining Network (CFM-Net) that employs an adaptive-depth feature extraction architecture to detect more precise and comprehensive artifact evidence across regions with heterogeneous AED values. Furthermore, we introduce an Orientation Guided Adversarial Training (OGAT) module to filter out identity information from PAD features while preserving the integrity of original artifact evidence. Experimental evaluations on LivDet datasets demonstrate the superior performance of our approach compared to state-of-the-art methods and achieve significant improvement in the classification ability of high AED fingerprints.
