Unideal Iris Segmentation Using Region-Based Active Contour Model

Robust segmentation of an iris image plays an important role in iris recognition. Most state-of-the-art iris segmentation algorithms focus on the processing of the ideal iris images that are captured in a controlled environment. In this paper, we process the unideal iris images that are acquired in an unconstrained situation and are affected severely by gaze deviation, eyelids and eyelashes occlusion, non uniform intensity, motion blur, reflections, etc. The novelty of this research effort is that we apply the modified Chan-Vese curve evolution scheme, which extracts the intensity information in local regions at a controllable scale, to find the pupil and iris boundaries accurately. A data fitting energy is defined in terms of a contour and two fitting functions that locally approximate the image intensities on the two sides of the contour. This energy is then incorporated into a variational level set formulation with a regularization term. Due to the kernel function used in energy functional, the extracted intensity information of the local regions is deployed to guide the motion of the contour, which thereby assists the curve evolution scheme to cope with the intensity inhomogeneity that occurs in the same region. The contours represented by the proposed variational level set method may break and merge naturally during evolution, and thus, the topological changes are handled automatically. The verification performance of the proposed scheme is validated using the UBIRIS Version 2, the ICE 2005, and the WVU unideal datasets.