Hand vein-based biometric authentication with limited training samples

[摘要] ENGLISH ABSTRACT : A number of novel hand vein-based biometric authentication systems areproposed. Said systems are non-intrusive and may for example assist withuser authentication at automated teller machines. An infrared image of eitherthe dorsal or ventral surface of an individual's hand is acquired through specialisedequipment, after which the geometrical properties of the hand are usedto extract a suitable region of interest (ROI). A novel protocol, which is basedon morphological reconstruction, is employed for the purpose of isolating theveins within the ROI. Feature vectors are extracted from the isolated veinsthrough the calculation of the discrete Radon transform. The feature vectorsare appropriately normalised in order to ensure rotational, translationaland scale invariance. The dissimilarity between the corresponding feature vectorsextracted from a questioned image and a reference image belonging tothe claimed client are represented by an average Euclidean or dynamic timewarping-based distance. A score-based or rank-based classi er is subsequentlyemployed for authentication purposes. It is demonstrated that, when only onetraining sample (of arbitrary quality) is available per client, and the client isgranted six opportunities for authentication, an average error rate (AER) of2.85% is achievable for a data set that contains dorsal hand vein patterns from100 individuals. In a scenario where the single training sample is guaranteedto be of very high quality and the client is granted only three opportunities forauthentication, the AER may be reduced to 0.77%.

[发布机构] Stellenbosch University