Cognitive Cyber Crimes in the Era of Artificial Intelligence

16Preventing Neural Data Leaks with Biometric Encryption

Antonio González-Torres

School of Computer Engineering, Costa Rica Institute of Technology, Cartago, Costa Rica

Abstract

The proliferation of neural networks in biometric systems has heightened concerns over neural data leaks, necessitating robust encryption mechanisms. This study introduces a novel approach to mitigating such leaks through biometric encryption, leveraging the BioDeepHash framework. Utilizing the SOCOFing dataset, comprising 6000 original and 49,270 synthetic fingerprint images, we implement a deep hashing technique that maps biometric data into stable codes, enhancing security and revocability. Our method achieves a genuine acceptance rate improvement of 10.12% for iris data and 3.12% for facial data compared to existing methods, with a false acceptance rate as low as 0% on the iris dataset and 0.0002% on the facial dataset.

Keywords: Biometric encryption, neural data leakage, deep hashing, SOCOFing dataset, BioDeepHash

16.1 Introduction

The integration of biometric [1, 2] systems with neural networks has revolutionized identity verification processes. However, the sensitivity of biometric data, coupled with the complexity of neural networks, introduces vulnerabilities to data leaks. Recent advancements, such as the BioDeepHash [3, 4] framework, address these concerns by mapping biometric data into stable codes using deep hashing techniques. Additionally, the use of convolutional neural networks (CNNs) ...