Circular and elliptical modeling for pupil boundary in closed-up human eye images

This paper proposes a method for detecting and modeling human pupil boundary in closed-up eye images. The proposed method is a model-based approach aiming to detect accurate pupil boundary in iris images. These images inhibit large variations such as different pupil size, shape and intensity background. The proposed method utilizes intensity and gradient information to detect edges of pupil boundary. These edges are often far from actual edges of pupil. Accurate edges of pupil are obtained by refined searching over a blurred input eye image using gradient information. The obtained edges are then modeled using circle and ellipse. From our experimental results, elliptical fitting yields better performance than circular fitting. A correct detection rate (CDR) of circular and elliptical fitting using CASIA-IRISV3-Interval database is 95.93% and 99.4%, respectively, whereas a CDR of circular and elliptical fitting using KSIP_DB01R database is 90.78% and 93.02%, respectively.