Multi-modal Finger Feature Fusion Algorithms on Large-Scale Dataset.

Multimodal biometrics technologies have become dominant in the domain of biometric recognition since they exploit complementary information between modalities and usually have better recognition performance. Thg finger has abundant biometrical features, including fingerprint, finger vein, and finger knuckle, which make it one of the most important research fields in multimodal biometric recognition. Though plenty of multimodal finger recognition algorithms have been investigated in literatures, most of them are based on two of three modalities of the finger. Besides, there is no open, simultaneously-collected, and large-scale trimodal finger dataset together with convincing benchmarks for scholars to learn and verify their multimodal finger recognition algorithms. In this paper, we propose a novel large-scale trimodal finger dataset containing fingerprint, finger vein, and finger knuckle. Two benchmarks from a feature-level fusion strategy and a score-level fusion strategy are established. Finally, comprehensive ablation studies are used to analyze the contribution of each finger modality.