End-to-End Automated Iris Segmentation Framework Using U-Net Convolutional Neural Network

Abstract

Iris segmentation plays an essential part in iris recognition because it defines the effective image region for the processes in iris recognition such as feature extraction and matching algorithm at later stages. However, the underlying algorithms for iris segmentation methods often involve heavy calculations and parameters, which are complicated, sensitive to noise and time-consuming. The robustness of the algorithms is often bounded by the assumptions made. The non-ideal conditions are the main culprit during iris segmentation process because the iris boundaries show insignificant variation which is indeterminate to the searching algorithms. With the success of deep learning models, researchers are actively looking towards convolutional neural networks (CNNs) to further improve on the accuracy of existing iris segmentation techniques. In this paper, we propose a simple and powerful deep-learning model for end-to-end iris segmentation based on the popular U-Net architecture. Our model is trainable end-to-end and, hence, avoids the need for hand designing the processing and segmentation procedures. Publicly available databases in visible light environment: noisy iris challenge evaluation part I and II (NICE1 and NICE2) and mobile iris challenge evaluation (MICHE-I) datasets were used in our experiments, promising performance has been demonstrated.