Global features for rapid identity verification with dynamic biometric data

Some of the biometrics used in identity verification, such as face, iris or fingerprint, have small fixed size pattern vectors which can easily be stored for rapid scoring. Important dynamic biometrics such as voice and signature, however, have much larger and variable sized feature vectors which require far greater verification computation. Such dynamic biometrics cannot therefore normally be used where verification must be performed rapidly on a device with little memory and very low speed. In this paper we compare the verification speed and accuracy obtained using different small fixed size global feature representations with that obtained, for voice and signature, using state of the art techniques with the full dynamic feature matrix. Using the best of these techniques we were able to reduce the multimodal processing time on a PDA SIM card, combining voice, signature and face biometrics, from one hour to ten seconds, while retaining a reduced but useful level of verification accuracy.