Iris recognition by fusing different representations of multi-scale Taylor expansion

The random distribution of features in an iris image texture allows to perform iris-based personal authentication with high confidence. We propose three new iris representations that are based on a multi-scale Taylor expansion of the iris texture. The first one is a phase-based representation that is based on binarized first and second order multi-scale Taylor coefficient. The second one is based on the most significant local extremum points of the first two Taylor expansion coefficients. The third method is a combination of the first two representations. Furthermore, we provide efficient similarity measures for the three representations that are robust to moderate inaccuracies in iris segmentation. In a thorough validation using the three iris data-sets Casia 2.0 (device 1), ICE-1 and MBGC-3l, we show that the first two representations perform very well while the third one, i.e., the combination of the first two, significantly outperforms state-of-art iris recognition approaches.

Research highlights
► Novel iris representations based on binary features from the multi-scale Taylor expansion.
► Enhancement of the local extrema-based approach with efficient matching.
► Combination of the above two performs with highest recognition rates.
► Evaluation results provided for each using Casia 2.0 (device 1), ICE-1 and MBGC-3l.