| Title |
Triplet Loss-based Finger-Vein Identification using Residual Networks |
| Authors |
강영묵(Youngmook Kang) ; 최정규(Jeongkyu Choi) ; 김학일(Hakil Kim) |
| DOI |
https://doi.org/10.5573/ieie.2021.58.5.31 |
| Keywords |
Finger-vein Identification; Deep learning; Triplet loss; CNN; Biometrics |
| Abstract |
The purpose of this paper is to develop a finger-vein identification algorithm by designing a deep neural network based on triplet loss for high accuracy. classifiers using deep neural networks require a lot of learning data to ensure high accuracy. However, in finger-vein identification, the classifier using the existing deep neural network is not suitable. Because the learning data is small and there are many classes to be identified. To solve this problem, we propose a network structure and loss function that extracts and identifies more differentiated features from finger-vein images for high accuracy. To learn the network by optimizing the characteristics of the finger-vein pattern, we construct a network that combines multi-class cross-entropy loss and triplet loss to effectively improve the identification accuracy. Experimental results show performance with 99.40%, 99.48%, 95.91% accuracy in the MMCBNU, FV_USM, and SDUMLA dataset. |