In order to make biometric systems faster and more user-friendly, lower-quality images must be accepted. A
major hurdle in this task is accurate segmentation of the boundaries of the iris in these images. Quite commonly,
circle-fitting is used to approximate the boundaries of the inner (pupil) and outer (limbic) boundaries of the iris,
but this assumption does not hold for off-axis or otherwise non-circular boundaries. In this paper we present a
novel, foundational method for elliptical segmentation of off-axis iris images. This method uses active contours
with constrained flow to achieve a simplified form of shape prior active contours. This is done by calculating a
region-based contour evolution and projecting it upon a properly chosen set of vectors to confine it to a class
of shapes. In this case, that class of shapes is ellipses. This serves to regularize the contour, simplifying the
curve evolution and preventing the development of irregularities that present challenges in iris segmentation.
The proposed method is tested using images from the UBIRIS v.1 and CASIA-IrisV3 image data sets, with
both near-ideal and off-axis images. Additional testing has been performed using the WVU Off Axis/Angle Iris
Dataset, Release 1. By avoiding many of the assumptions commonly used in iris segmentation methods, the
proposed method is able to accurately fit elliptical boundaries to off-axis images.