On the intra-subject similarity of hand vein patterns in biometric recognition

• Published in: Expert systems with applications Vol. 192; p. 116305
• Main Authors: Kuzu, Rıdvan Salih, Maiorana, Emanuele, Campisi, Patrizio
• Format: Journal Article
• Language: English
• Published: New York Elsevier Ltd 15.04.2022; Elsevier BV
• Subjects: Biometric recognition; Biometrics; Deep learning; Feature extraction; Fingers; Hand (anatomy); Machine learning; Palm; Pattern recognition; Similarity; Vein patterns; Deep learning; Biometric recognition; Vein patterns
• DOI: 10.1016/j.eswa.2021.116305
• ISSN: 0957-4174

In these years, biometric recognition based on hand vein patterns is receiving an always increasing attention from both industry and academia, thanks to the advantages it offers with respect to conventional approaches, such as those relying on fingerprint, iris, or face. Nevertheless, there are still several properties of vein traits that need to be investigated and well understood. In this paper, we here analyze the level of similarity, evaluated in terms of recognition rate of a biometric system, of vein patterns in the fingers, palms, and dorsa of a person’s left and right hands. In other words, we analyze whether a subject, enrolled using vein patterns, either finger-vein, palm-vein, dorsal-vein, from one hand, can be recognized using the corresponding patterns from the other hand. Our investigation is conducted using deep-learning-based feature extraction approaches, three different vein modalities, and four different databases. The obtained experimental results show that corresponding fingers, palms, and dorsal regions from different hands of the same subject show more resemblance with respect to the traits from the same hand of different persons. Furthermore, our findings point out that similarities among vein patterns in corresponding fingers could be used for recognition purposes, while this still cannot be applied to palm and dorsum vein patterns. •Similarity between left and right hand vein patterns analyzed in a biometrics context.•Evaluation conducted on palm, dorsum, and finger veins from four different datasets.•State-of-the-art convolutional network employed for feature extraction.•Different network training models used to analyze vein pattern cross-hand similarity.•Activation maps computed to show discriminative patterns in distinct training models.