Noninvertible Transformation schemes for face template protection

Biometric encryptions combine biometric authentication and cryptography. It can possibly provide accurate and convenient authentication with the benefits of privacy and security. In this paper, we propose three irreversible transformation schemes using four irreversible transformation functions for face features: order number transformation scheme, feature value transformation scheme and two dimensional transformation schemes. In experiment，we compared the authentication performance and the security of each scheme. Faces from ORL database are taken as examples to illustrate the performance. Then we use one arbitrary face from the database to compare the reconstructed faces for each transformed feature. The results show that the reconstructed face images from the transformed feature values are much different from the original face images. The scheme 2, order number transformation, can be secure enough by increasing the percentage of many-to-one area. It has almost the same authentication performance as original face features.