Privacy Protection of Biometric Templates Using Deep Learning

With the growing use of biometric authentication in real-world applications, there are growing concerns related to the privacy of biometric templates enrolled in the database. Many algorithms have been proposed for the protection of biometric templates, however, this leads to a trade-off between the security of templates and matching performance. In this work, we propose a robust framework for improving upon the existing biometric template protection techniques. The proposed method uses one-shot enrollment for the mapping of images against unique binary codes assigned to them. The binary codes are obtained by binarization of image pixels. This is followed by the use of the SHA-256 hash function, and the class labels are generated based on the hashes templates. This is followed by the extraction of features using a deep convolutional architecture. During verification, the predicted labels are matched against labels generated from the hashed templates of the users stored during enrollment. The proposed framework is capable of achieving state-of-the-art performance.