A Framework for Iris Localization based on Greedy Snake Model

Abstract

This paper proposes a framework for iris localization with greedy snake model to accurately
extract iris region and compensate for the accuracy problem caused by the non-standard circle
characteristics of the iris. Initially upper eyelashes are detected and removed with mathematical
computational basis operators within a windowed eye region. In the proposed scheme, the non circular
pupil contour is detected in an iterative fashion with a novel edge based two stage greedy snake model. In
the first stage, the pupil-iris edge is coarsely located with mathematical computational gradient detector and
in the second stage, the precise pupil contour is detected with greedy snake in which the contour is
initialized within the pupil and deformed into new shape in response to the two controlling force models,
introduced as internal and external forces to properly activate the contour. The image gradient and the
curvature are utilized together to determine the speed and direction of the contour deformation,while for the
localization of limbus boundary vertical edges are detected between iris and sclera region with horizontal
polynomials coefficient. Then the precise limbus boundary is localized from the two annulus sector area
with the detection of radial boundary points in a sequence along angular directions within the specified
projection curve radiating from pupil center. The experimental results with standard CASIA database show
that the proposed scheme is robust in finding exact noncircular pupil, limbus boundary and eyelids.