Abstract:

An iris biometric system has a versatile applicability. At this point, there is a wide number of areas where iris biometrics is successfully implemented as a means of access, areas such as finance, banking, healthcare, welfare, border control, point of sale and ATM’s. The performance of an iris biometric system can be sensible to any error that may perpetuate, starting even from the eye image acquisition. It is obvious that an iris biometric system that functions with errors would eventually lead to behaving in an undesired manner: consistently rejecting its users and/or authorizing persons from outside the system. This article has the purpose of identifying the influence of one particular error: the Eccentricity Detection Error (E.D.E.), a newly identified type of iris segmentation error. Having a non-zero E.D.E. means having, for two iris images that arc acquired from the same person, a different distance between the two vectors determined by the pupillary (inner) boundary centers and the limbic (outer) boundary centers. This distance is established on the basis of the approximated (during segmentation) inner and outer boundary centers for the two irides. This article presents the results of iris recognition using a PNN neural network, performed with iris codes at different values of E.D.E.: O, lower than 2, and larger than 2. The databases used for the experimental tests are ND-Crosssensor-lri5Z0l3
(LG2200), CASIA Lamp (V4), and CASIA Interval (V4).