An Improved Finger Vein Recognition Model with a Residual Attention Mechanism

• Published in: Biometric Recognition Vol. 12878; pp. 231 - 239
• Main Authors: Liu, Weiye, Lu, Huimin, Li, Yupeng, Wang, Yifan, Dang, Yuanyuan
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2021; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Attention mechanism; Biometrics; Deep learning; Finger vein recognition; Inception module
• ISBN: 3030866076; 9783030866075
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-86608-2_26

Deep learning-based Biometric authentication has become one of the most popular research subjects in the field of Computer Vision. In this paper, we propose a novel model architecture for finger vein recognition based on an improved residual attention network. First, we squeeze the size of the original network to adapt to the training data scale. Then, to prevent excessively repeated operations of linear extraction, we introduce the Inception unit to replace some residual units in the original model. The multi-branch structure can learn vein features from different aspects. Besides that, with the attention block, primary vein patterns can be extracted and the bottom-up, top-down structure activates feature maps with learned attention weights. The experimental results show that our model acquires 98.58% and 97.54% accuracy on two public datasets, respectively. Compared with state-of-the-art models, the proposed model has fewer parameters and better performance.