Face Recognition and Template Protection with Shielding Function

Abstract

Biometric authentication is the use of unique human features to provide a secure, reliable and convenient access to an environment or a computer system. However, there are numerous security and privacy concerns associated with the use of biometrics as a means of authentication. Unprotected biometric data can be used by an impostor to impersonate legitimate uses, to violate their privacy and steal their identity. This paper proposes a simplified, secure and privacy-preserving authentication scheme for face biometric based on modified shielding function. The modified shielding function is a simplified version of the generic shielding function which does not require additional preprocessing steps of quantization and reliable bit selection. Rotation invariant neighbour-based local binary pattern (RINLBP) is used to extract fixed length binary features directly from pre-processed face images. RINLBP is simple to calculate and has good performance. It is also robust against changes in illumination and image rotation. Concatenated error correction technique is used to address errors due to noise and intra-class variation. The concatenated technique corrects errors both block and bit errors in contrast to the generic shielding function in which only bit level errors are corrected. Results of experiments based on 200 face images obtained from the CASIA near infrared face database show a false acceptance rate of 0.47% and a false rejection rate of 1.56%. Our scheme has a key length of 120 bits, which is higher than the minimum requirement of 50 bits for biometric keys. It also has a large key space and entropy which makes it less susceptible to guessing attack (Pr =0.008).