Effects of enrollment templates count on iris recognition performance using reliable bits

Biometric authentication is among the most convenient, reliable and secure techniques for human identification and recognition. In general, iris-based biometric systems provide very good results in terms of recognition accuracy but, as the noise in input images increases, the performance of these systems starts to fall down. Using multiple templates at enrollment time can be very effective for improving recognition accuracy in noisy iris image datasets. In this paper, we introduce techniques to generate a more reliable final template from multiple enrollment templates. Our experiments with CASIA 1, CASIA 3 and BATH iris image datasets show that recognition performance is significantly improved using the proposed techniques.