Iris recognition by local extremum points of multiscale Taylor expansion

Rich texture of iris image allows to perform iris-based person authentication with very high confidence. We propose to use the most significant local extrema of the first two Taylor expansion coefficients as descriptors of the iris texture. The proposed features can be efficiently compared and even can correct moderate inaccuracies in iris segmentation during the matching stage. A brief introduction of our iris segmentation algorithm is followed by the analysis of the proposed features and the similarity function. We provide experimental results of verification quality for three commonly used iris data sets. An analysis of strong and weak aspects of the proposed approach is done.