Biometric Authentication with Combined Iris and Sclera Information by YOLO-based Deep-Learning Network

In this work, the effective YOLO-based iris and sclera deep-learning classifier is studied for biometric authentication. By joint the partial iris and sclera information, the traditional sclera and iris segmentation pre-process is not required. By the YOLO-based model, the visible-light eye images are labelled with the jointly partial iris and sclera region, and the identity classifier is trained to inference the person's identity. By the UBIRIS image database for performance evaluations, the proposed YOLOv2 based joint iris and sclera identity classifier achieves the mean average precision (mAP) up to 99.83%, and the average inference time per one image of the YOLO model can be also reduced by using less number of anchor boxes. Compared with the previous works, the proposed design is more effective without any iris and sclera segmentation computations.