Pupil Boundary Detection for Iris Recognition Using Graph Cuts

In this paper an automatic segmentation method for accurate pupil boundary detection for iris recognition purpose using graph cuts is presented. Most iris segmentation algorithms used for iris recognition consider the pupil as a circular area and fit a circle to its boundary. Considering the pupil to be a circle is a method which is sensitive to the imaging condition because in the case of off angle imaging the pupil becomes an ellipse instead of a circle. In addition the pupil cannot be considered a perfect circle even if the off angle imaging is avoided. Most information in iris area exist in the collarette which is a small area around the pupil therefore a small error in detecting the pupil boundary may result in poor performance of the identification system. The presented graph cut method uses the gray level values of pixels to compute the weights of the links in the graph. The graph has two terminals, one represents the pupil and the other one represents the rest of the image, considered as the background. The method is explained and its high performance is demonstrated by experiments based on applying the method on the CASIA eye image database.