Deep Iris Feature Extraction
Andrej Hafner, Peter Peer, Žiga Emeršič, Matej Vitek
Abstract
Iris recognition refers to the automated process of individual recognition based on the patterns in their irises. Due to its uniqueness, it is a common modality used in biometric recognition. With a technique pioneered by Daugman, it was shown that it enables recognition with very low false match rates. However, existing approaches still offer room for improvement in terms of accuracy. To address this, we adapt the pipeline defined by Daugman using convolutional neural networks to function as feature extractors and train the convolutional neural networks on a part of CASIA-Iris-Thousand dataset for closed set prediction. Trained models are then used for feature extraction, enabling us to perform open set recognition. With DenseNet-201 we achieve 97.3% recognition accuracy in closed set recognition and 98.5% recognition accuracy in open set recognition, achieving state-of-the-art results.