Vote-based Iris Detection System

Abstract

Finding the accurate location of iris is crucial to some applications in biometrics, human computer interaction and medical research. The accuracy of the location will affect the outcome of following iris segmentation, features extraction and measurement, to name a few. This paper presents an accurate vote-based method to detect and localize both irises from color images. The algorithm starts with image filtering steps such as Gaussian filtering to reduce the effect of various lighting conditions. Then, iris candidates will be generated after the detection of reflection in iris. A cost will then be computed for each iris candidate according to the contribution from generic eye template, intensity variation factor, circularity factor and reflective factor. Finally, a pairing process is used to determine the real iris pair in order to locate both irises. Our experiment on Michigan database has reported a promising accuracy of 91.21%.