Efficient feature matching in a very large iris database for person identification

In this paper, a new efficient feature matching method for a very large iris database is proposed. The new method is particularly useful for the iris recognition system that works with the popular IrisCode features. The method initially performs a partial feature matching between segments of IrisCodes after random permutation. This partial matching results in a reduced set of candidate IrisCodes on which complete matching is then performed. Both the partial and complete matching are performed by setting decision thresholds for the hamming distances computed between IrisCodes. The results of performance measures such as the hit rate and computational complexity reduction rate show the effectiveness of the new method in searching a very large database. The method can be easily extended to similar high dimensional binary pattern matching problems such as audio fingerprinting.