Preventing Neural Data Leaks with Biometric Encryption

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.