Design of Low-Complexity YOLOv3-Based Deep-Learning Networks with Joint Iris and Sclera Messages for Biometric Recognition Application
Chia-Wei Chuang, Chih‐Peng Fan, KyungHi Chang · 2020
In this study, the effective low-complexity YOLOv3_tiny based deep-learning inference networks are studied for biometric authentication. First, the eye images are labelled by jointly partial iris and sclera zones, and then the proposed YOLOv3_tiny based classifier infers the person's identity efficiently. By the UBIRIS database, the applied YOLOv3_tiny based inference model achieves the mean average precision (mAP) up to 99.92% with only using one anchor box. Compared with the previous ocular biometric studies with iris or sclera information, the proposed low-complexity design provides better performance of accuracy and does not need the iris and sclera segmentation process.