Encoder-Decoder Convolutional Neural Network based Iris-Sclera Segmentation

Abstract

Iris-sclera biometry is one of the features that yields high accuracy in user recognition and liveness detection systems. In this study, segmentation processes the first stage of an iris-sclera user verification system have been considered. Traditional and convolutional neural network based deep learning methods have been used for iris-sclera segmentation. Performance of the investigated methods has been tested on two distinct eye image datasets (UBIRIS and self-collected data). Our experimental results show that deep learning based segmentation methods outperformed conventional methods in terms of dice score on both datasets.