Robust pupil boundary detection by optimized color mapping for iris recognition

Accurate pupil segmentation is the first and most important step for an iris recognition system. Current methods are based on fitting a model such as circle or ellipse to find and detect pupil, while these methods don't have sufficient accuracy and are sensitive to the specular spot reflection. In this paper, we utilize an optimized color mapping to increase the accuracy of pupil segmentation, regardless of pupil model and its shape (circular or elliptic), while removing the effects of specular spot reflection. The optimum color mapping can be established by an iterative minimization algorithm similar to Levenberg-Marquardt (LM) method. By applying this method, a new image is provided with a clear pupil region that can be easily segmented. Also a robust preprocessing step is presented in this paper that sharpens and clears pupil region. We obtain 98% accuracy in pupil boundary detection by applying this method on UBIRIS dataset. Also, the proposed method works well on any model of eye image even where the eye is not perpendicular to the camera.