Iris Feature Extraction Using Optimized Method of Selecting Points with Less Occlusion

Abstract

In Daugman's iris recognition method, the application points determine which pixels of the normalized iris images will be used in the matching stage of the algorithm. In his work, those points are chosen in an equidistant form, referenced here as homogeneous distribution. The homogeneous distribution of these points, often selects pixels that represent eyelids, eyelashes and specular reflections, occlusions that should be extracted from the matching step. A binary mask (occlusion mask), in the matching step, enables disregarding the computation of these bits. However, some template protection schemes have restrictions on the use of such masks, either because of memory/computational cost limitations or because of limitations of the algorithm itself. In this paper, we propose a method that optimizes the distribution of the application points avoiding regions with high rate of occlusions, reducing the impact of not using the occlusion mask in the matching step. The method is based on statistical analysis. The new application points distribution is called optimal distribution. The recognition performance obtained with the optimal distribution of the application points was EER = 3.1% and F RR = 6.3% (for F AR = 0.1%) while for the homogeneous distribution without the usage of masks EER = 4.8% and F RR = 12.7% (for F AR = 0.1%).