Iris matching using ferns classifier
Lu Chen, Shu Zhang, Mei Hua Xie · 2012
In conventional iris recognition Systems, iris matching is considered as binary-classification problem-authentic matching and imposter matching. Many existing methods employ simple distance to implement iris matching, such as Hamming distance. These methods can't make full use of iris feature and generate relatively high FRR (False Reject Rate), FAR (False Accept Rate), in addition, these methods are less robust. To overcome this problem, in this paper, we treat iris matching as a multi-classification problem and use ferns classifier which fit nicely into a naive Bayesian framework to implement iris matching. To recognize iris, the classifier uses hundreds of binary feature vectors and approximates class posterior probability. Finally, our proposed iris matching method is tested in verification experiment. Experimental results show that the proposed method can greatly improve performance of iris recognition system.