Biometric Cryptosystem with Deep Learning: A New Frontier in Security

Abstract

The Biometric cryptosystem uses a variety of methods to protect templates. In this work, a deep learning-based approach to improve the robustness of the fuzzy vault scheme in biometric cryptosystems. Our approach uses a CNN to extract distinctive features from biometric data and generate the polynomial equation that unlocks the vault. We evaluate our approach on a dataset of fingerprint images and demonstrate that it achieves higher accuracy of 89.9% than traditional methods. The relation between original and decrypted image is computed based on various parameters such as Cr., MSE, MAE etc. and overall fair performance is achieved on four fingerprint databases.