Modeling the individuality of iris pattern and the effectiveness of inconsistent bit masking strategy

Iris recognition is one of the most accurate biometric technologies. The uniqueness of iris, also known as iris individuality, has been widely accepted as one foundation for iris recognition. Although a few iris individuality models have been proposed, they are either incomplete or less accurate. In this paper, we investigate the iris individuality problem using Daugman's iris code method. We divide the bits in an iriscode into two groups, i.e., consistent and inconsistent bits, and provide the individuality analysis by both FAR and FRR modeling. Numeric evaluation using real iris data shows its usefulness in predicting the empirical performance. Furthermore, till now it is just experimentally confirmed that the recognition accuracy could be improved by masking out inconsistent bits. In order to formally e- valuate the effectiveness of this strategy, we derive the iris individuality model after masking out the inconsistent bits. Comparison of the two models has demonstrated the improved accuracy of the masking strategy, and the drop of EER is up to about 80%.