A finger-vein recognition method based on double-weighted group sparse representation classification

Finger-vein recognition is a new type of biometric identification technology compared to traditional biometric identification methods such as fingerprint recognition. Finger-vein is considered a safe and reliable biometric pattern due to its vivo nature and has been widely used in various fields. However, finger-vein images are susceptible to finger thickness and near-infrared light distribution during the process of acquisition, which often results in low-quality images (e.g., low contrast and overexposure), causing feature loss and affecting the final recognition accuracy. In this paper, a method based on double-weighted group sparse representation classification is proposed to improve the recognition performance for low-quality finger-vein images. The proposed method represents the test sample by a sparse linear combination of the training samples while making full use of those training samples’ data and label information through the group sparsity constraint. In addition, considering the large intra-class differences and small inter-class differences in low-quality finger-vein images, weight constraints for sparse coefficient vectors are added at the individual and group levels to reduce the impact of low-similarity training samples and heterogeneous training samples on the final recognition results, respectively. Compared with the state-of-the-art methods on three public datasets, experimental results demonstrate that the proposed approach achieves better recognition accuracy and robustness, especially for low-quality finger-vein images.