A comprehensive evaluation of deep learning models for fingerprint liveness detection

Abstract

Fingerprint recognition systems have shown some weaknesses related to security issues such as fake presentation attacks. Therefore, it is necessary to protect these systems against attacks by incorporating fingerprint liveness detection (FLD) algorithms that must be able to distinguish between live and fake fingerprints. In this work, we propose a deep learning framework for FLD. Specifically, deep neural networks based transfer learning is proposed to classify the fingerprint as live or fake fingerprint. The proposed method focuses on the important features of fingerprint, which allows enhancing the feature representation and suppresses the less relevant ones. We evaluated the proposed FLD on LivDet2023 database, and results, meticulously analyzed, unveil the superior performance of the proposed method. Notably, the proposed method with DensNet201 model attains an exceptional accuracy of 98.36% on the LivDet2023 dataset, surpassing other deep networks