Abstract: Obtaining the iris localization precisely and fast is the prelude of effective iris recognition. Traditional iris localization methods, including Daugman localization and Hough transformation localization, are weak in processing images with thick eyelashes and severely shlter. In this paper, combined with previous work of other researchers, deep learning method was employed to classify iris images based on the characteristics of iris region. We carried our experiment on the CASIA-IrisV3-Interval dataset, to verify the effectiveness of our work. The accuracy of pixel-wise classification is around 98.4%, with a higher robustness.