Palm Vein Recognition with Deep Hashing Network.

Human biometrics has strong potential of robustness, safety and high authentication accuracy. As a new biometric trait, palm vein recognition attracts spacious attention nowadays. To further improve the recognition accuracy, we propose an end-to-end Deep Hashing Palm vein Network (DHPN) in this paper. Modified CNN-F architecture is employed to extract vein features and we use hashing code method to represent the image features with a fixed length binary code. By measuring the Hamming distances of two binary codes of different palm vein images, we can determine whether they belong to the same category. The experimental results show that our network can reach a remarkable EER = 0.0222% in PolyU database. Several comparative experiments are also conducted to discuss the impact of network structure, code bits, training test ratio and databases. The best performance of DHPN can reach EER = 0% with 256-bit code in PolyU database, which is better than the other state-of-art methods.