An Iris Recognition System Using A New Method of Iris Localization
Ahmed Ak. Tahir, Sarhan S. Dawood, Steluta Anghelus · International journal of open information technologies · 2021
Abstract —An iris recognition system for person identification is developed with a new method for iris localization. For pupil boundary detection, a method robust to the specular point reflection problem is developed. It consists of a morphological filter and two-direction scanning methods. For limbic boundary detection, the Wildes method is modified by restricting the process of Canny edge detector and Hough transform to a small Region-Of-Interest (ROI) not exceeding 20% of the image size. For eyelid detection, the method of Refine-Connect-Extend-Smooth (R-C-E-S) is used, which detects three possible cases (single eyelid, both eyelids, and free iris). For iris normalization, rubber-sheet model transform is used and for iris coding, the Gabor filter is used. The performance of the system is evaluated for the individual stages and for the whole system using three different databases (CASIA-V1.0, CASIA-V4.0-Lamp, and SDUMLA-HMT). The accuracy of correct detection reached 99.9%-100% for pupil boundary and 99.6%-99.9% for limbic boundary detection. For eyelid detection; the accuracy reached 93.2%-97.6% for the upper eyelid, 95.3%-99.15% for the lower eyelid, and 96.7%-96.92% for free iris (iris not occluded by eyelids). The overall accuracy and the Equal Error Rate (EER) of the system for the CASIA-V1.0 database are 96.48% and 1.76%, for CASIA-V4.0-Lamp, are 95.1% and 2.45%, and for SDUMLA-HMT are 93.6% and 3.2%.