One-way randomised radon mapping for appearance-based biometric authentication

Biometrics has become one of the strongest replacements for tradition authentication systems. However, in order to replace the password/token system with a biometric one, biometric data has to be fully protected at all stages of processing and authentication. In this paper a new cancellable biometric transformation is presented. The Randomised Radon Signature (RRS) cancellable templates of face biometrics are generated by extracting a set of Radon signatures in the signal domain of the face image. This extraction is governed by a randomised weighting function. The extracted signatures are then mixed with a Random projected face image to lock the template. The RRS templates can be easily regenerated by using different random subspaces of the weighting function and the projection matrix. They can also be used in the same way as original images for authentication in statistical holistic approaches of face recognition. The security of the system has been analysed by a derivation of the guessing entropy of the RRS templates. Using the fisherface algorithm for evaluation, the accuracy performance has been improved by 100% for the RRS over the original images since the genuine and impostor distribution separation has been increased by 261.5%, consequently this has given rise to a dramatic reduction in the equal error rate.