1. Abstract—In this paper, we propose a revocable and privacy preserving template of face biometrics based on random projection. The face biometric is augmented and simultaneously projected onto random subspace. The face image vector is augmented by adding the vector whose elements are varying with zero mean. The augmented face vector can provide better accuracy of authentication and privacy preservation. We analyze the similarity, privacy preserving and security properties of the proposed augmented face biometric information in the random projection-domain. To demonstrate the feasibility of the proposed method, detailed theoretical analysis and several experimental results are provided. The results show that our method is able to provide revocability and privacy preservation while offering better authentication accuracy and security. I. INTRODUCTION ONVENTIONAL identity authentication mechanism is based on cryptographic key such as AES, DES, and RSA. Although cryptographic key-based authentication guarantees secure protection of the information, it has a difficulty in key management: The keys are so long that it is impossible to memorize. Usually the keys are stored securely and released based on alternative authentication mechanism. The alternative authentication is based on knowledge or possession. The commonest knowledge based authentication mechanism is a password based cryptosystem. In the password based cryptosystem, the security depends on the complexity and the size of the password: A short password compromises the security while a long password is difficult to remember. In the case of possession based authentication, possessions such as tokens, ID cards etc, are used to carry the secret key. However, it cannot guarantee security since it can be lost and stolen. Moreover, neither of cryptosystems can provide non- repudiation. On the other hand, biometric-based authentication systems confirm an individual’s identity based on the physiological and/or behavioral characteristics of each user. Therefore, there is nothing to lose or forget, since each user does not have to carry or memorize any authentication key. These systems are also relatively difficult to circumvent. Although the biometric-based authentication mechanisms provide various advantages, there still exist major problems in biometric based authentication systems: revocability, privacy, and accuracy [1]. Biometric data has very unique Manuscript received June 1, 2009. H. Sohn and Y. M. Ro are with the Department of Electrical Engineering, Korea Advanced Institute of Science and Technology, Daejeon, Republic of Korea (e-mail: sohnhosik@kaist.ac.kr, ymro@ee.kaist.ac.kr ) K. N. Plataniotis is with the Edward S. Rogers Sr. Department of Electrical and Computer Engineering, University of Toronto, Toronto, Canada (email: kostas@comm.utoronto.ca ) characteristics that represent an individual’s identity. Thereby, biometrics cannot be easily cancelled and reissued due to the limited number of biometric traits that a human has. Also, the uniqueness of biometric contains potential risk of privacy invasion. Since biometric data reflects the user’s physiological (e.g., face, fingerprint, etc.) or behavioral characteristics (e.g., voice, signature, etc.), user’s privacy can be revealed if the templates are compromised. Moreover, biometric authentication is based on fuzzy match due to the noisy nature of biometric data, while exact match can be obtained in cryptographic key-based authentication. Several efforts have been dedicated to the generation of revocable and privacy preservable biometrics. One well-known solution is to bind biometrics with random cryptographic keys [2]. Soutar et al [3] presented a correlation-based method for fingerprint verification, and Davida et al [4] proposed to store a set of user specific error correction parameters as template for an iris based system. In [5], Juels and Wattenberg introduced an error correction based method, fuzzy commitment scheme, which generalized and improved Davida’s methods. Later, a polynomial reconstruction based scheme, the so-called fuzzy vault, has been proposed [6], [7]. In [8], Boult et al. proposed a robust distance measure that improves the accuracy of face recognition while supporting revocability of biometric tokens. Recently, Teoh et al [9] introduced a BioHashing method which produces revocable and non-invertible biometric template. In [10], multi-space random projection is applied to generate privacy preserving and cancellable biometrics. The major challenge of these approaches lie on authentication accuracy and error tolerant capability to retrieve the key from noisy biometric data. In this paper, we propose a method to generate revocable and privacy preserving template of face biometrics based on random projection. The proposed scheme is a two-factor authenticator that utilizes face biometrics and user-specific key. The key is used to generate a random matrix for projecting face image vector onto random subspace [1]. A face image vector is augmented and simultaneously projected onto random subspace. In the proposed method, the augmented face is generated by increasing the variation of elements in the vector. The augmented face vector provides better accuracy of identity authentication and privacy preservation. We compare the proposed method with the method [1] of translating the mean of face vector. We show that the proposed augmentation of face vector provides stronger security. To demonstrate the efficiency of the Biometric Authentication using Augmented Face and Random Projection Hosik Sohn, Yong Man Ro, Senior Member, IEEE, and Konstantinos N. Plataniotis, Senior Member, IEEE C 978-1-4244-5020-6/09/$25.00 ©2009 IEEE