DBANet: A Dual Branch Attention-Based Deep Neural Network for Biological Iris Recognition
Jun Chen, Yangguang Cui, Fuke Shen, Jianhua Shen, Tongquan Wei · 2022 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) · 2022
Emerging iris recognition techniques are highly dependent on high-resolution iris images. However, existing iris recognition methods cannot effectively extract local texture features in low-resolution application scenarios, resulting in low recognition accuracy below expected. In this paper, we propose DBANet, a novel dual branch attention-based deep neural network for biological iris recognition that can achieve high accuracy for both high and low-resolution images. Specifically, we first design a spatial feature module with a small stride to preserve lower-level spatial detail features. Then, since the high-level feature can provide rich global context information, we propose a context feature module to generate high-level features. Finally, we develop a novel spatial attention module to fuse features generated by the above modules. We conduct the experiments on UBIRIS. v2, CASIA-V4-Distance, and MICHE-I datasets. Experimental results show that as compared to state-of-art methods, our proposed method can reduce equal error rates by up to 38.7%, 53.4%, and 71.9%, respectively.