A Novel Fingerprint Biometric Cryptosystem Based on Convolutional Neural Networks

Časopis: Mathematics

Volume, no: 9 , 7

ISSN: 2227-7390

DOI: 10.3390/math9070730

Stranice: 1-12

Link: https://www.mdpi.com/2227-7390/9/7/730

Modern access controls employ biometrics as a means of authentication to a great extent. For example, biometrics is used as an authentication mechanism implemented on commercial devices such as smartphones and laptops. This paper presents a fingerprint biometric cryptosystem based on the fuzzy commitment scheme and convolutional neural networks. One of its main contributions is a novel approach to automatic discretization of fingerprint texture descriptors, entirely based on a convolutional neural network, and designed to generate fixed-length templates. By converting templates into the binary domain, we developed the biometric cryptosystem that can be used in key-release systems or as a template protection mechanism in fingerprint matching biometric systems. The problem of biometric data variability is marginalized by applying the secure block-level Bose–Chaudhuri–Hocquenghem error correction codes, resistant to statistical-based attacks. The evaluation shows significant performance gains when compared to other texture-based fingerprint matching and biometric cryptosystems.
Ključne reči: biometric cryptosystem; fuzzy commitment scheme; fingerprint recognition; machine learning; convolutional neural network