Pupil Segmentation Using Orientation Fields, Radial Non-Maximal Suppression and Elliptic Approximation

This paper proposes a novel pupil segmentation method for robust iris recognition systems. The proposed method uses orientation fields to accurately detect an initial pupil center, and applies radial non-maximal suppression to remove non-pupil boundaries. Finally, we repeatedly fit the pupil boundary by radius-updating, center-shifting and region of interest (ROI) shrinking adjusting the radius and center of a circular model, and the estimated pupil boundary is approximated with a novel elliptic model. By the elliptic approximation, the pupil boundaries are more correctly segmented than those of circular models. The detection hit ratio is largely improved due to robust detection of the initial centers. The experimental results show that the proposed method can accurately detect pupils for various iris images.