Abstract
In this thesis, we present a novel statistic approach, named probabilistic feature extraction (PFE), for human iris. All the templates of each subject are considered in order to obtain statistical properties for personal iris data. In this thesis, we use CASIA-V2 database to do training and build confirmative templates for authentication system. We use binary and real-valued iris templates on our approach. In addition, two concatenated schemes are investigated to further enhance the performance of authentication. Firstly, error correcting code (ECC) is an algorithm to correct error bits which occur during the transmission through communication channels. The templates in enrollment and authentication stages can be regarded as the messages before and after the transmission, respectively, and ECC can be applied to enhance the authentication reliability. Secondly, hidden Markov model (HMM) is a statistical model in which the system being modeled is to be a Markov process with unobserved states. We attempt to achieve iris pattern authentication based on the use of HMM model. Especially, we also apply real-valued templates on ECC soft decoding. All the proposed iris authentication systems measure the equal error rate (EER) for evaluating the performance. By using binary iris templates, our proposed method is about 0.25%. If ECC and HMM are applied, the ERR is about 0.017% and 0.2%, respectively. On the other hand, by using real-valued iris templates, our proposed method is about 0.07%. We apply ECC soft decoding and the ERR is about 0.005%.