Whether normalized or not? Towards more robust iris recognition using dynamic programming

Abstract

Iris recognition is one of the most promising fields in biometrics due to more accurate, convenient and low-cost. However, it is still a challenging task for application in practical complex scenarios. More attention have been paid on non-ideal iris segmentation and cross-system feature extraction in recent years. In order to solve the issues, this paper investigates a novel non-normalized preprocessing method based on dynamic path search for iris segmentation. Meanwhile, we employ a deep convolution network (DCNN) based on partial convolution operators to extract iris features. Through benchmark experiments on two public iris datasets CASIA-Iris-Thousand (CASIA) and IIT Delhi Iris Dataset (IITD), we achieve the significant and encouraging results, which demonstrate the effectiveness of the proposed methods. More importantly, we prove that using iris segmentation images without normalization may be a better choice when exploring iris recognition solutions based on deep learning.