Robust distance measures for face-recognition supporting revocable biometric tokens

This paper explores a form of robust distance measures for biometrics and presents experiments showing that, when applied per "class" they can dramatically improve the accuracy of face recognition. We "robustify'' many distance measures included in the CSU face-recognition toolkit, and apply them to PCA, LDA and EBGM. The resulting performance puts each of these algorithms, for the FERET datasets tested, on par with commercial face recognition results. Unlike passwords, biometric signatures cannot be changed or revoked. This paper shows how the robust distance measures introduce can be used for secure robust revocable biometrics. The technique produces what we call Biotopestrade, which provide public-key cryptographic security, supports matching in encoded form, cannot be linked across different databases and are revocable. Biotopes support a robust distance measure computed on the encoded form that is proven not to decrease, and that may potentially increase, accurately. The approach is demonstrated, to improve performance beyond the already impressive gains from the robust distance measure