Fingerprint Anti-spoofing Analysis: From Minutiae to Transformers

In the contemporary digital and interconnected world, the need for robust security measures is paramount. From personal devices to national security, biometric recognition has enhanced security, streamlined authentication processes, and improved overall efficiency. Fingerprint liveness detection plays a vital role in ensuring the security and reliability of biometric systems by differentiating between genuine fingerprints and presentation attacks. This work presents a detailed analysis on various computational algorithms for fingerprint liveness detection along with cross-sensor data evaluation. The study investigates the effectiveness of different feature extraction modules, including local binary pattern (LBP), histogram of oriented gradients (HOG), scale-invariant feature transform (SIFT), minutiae points, and deep feature extractors, including various convolutional neural architectures and transformers. Additionally, deep learning patch-based classification approaches with multiple weight initialization strategies and transfer learning from large-scale datasets are explored for liveness detection. The overall framework is achieving an outstanding average classification error (ACE) value of less than 0.5%. Moreover, the comparison of different architectures, evaluation on various large datasets, model complexity versus anti-spoofing robustness analysis and cross-sensor evaluations offer useful insights about fingerprint liveness detection.