An Iris localization method for noisy infrared iris images

Iris segmentation is an important step in iris biometrics and becomes more challenging when iris images are noisy. Iris segmentation process localizes iris in the iris image and detects eyelids also. The proposed method localizes iris in noisy frontal view iris images captured under near infrared (NIR) illumination and having noise issues such as lighting and specular reflections, eyeglasses, low contrast, non-uniform illuminations and occlusions by heavy eyebrows, eyelids, eyelashes and hair. The proposed method is based on circular Hough transform (CHT) and an integro-differential operator (IDO) derived by taking motivation from Daugman´s IDO. The pupil is localized using image binarization followed by CHT based algorithm and iris´s outer boundary is detected by searching a limited image domain for maximum gray difference between iris and sclera using the proposed IDO. The method was tested on noisy images from two public NIR iris databases: MMU V2.0 and CASIA-Iris-Thousand V4.0. The average accuracy of the proposed method is 99.05 % and average time cost per image is 415 ms. Comparison of results of the proposed method with some published state-of-art iris localization methods proves its novelty.