Deep-Wrist: Reliable User Verification using Deep Representation of WristVein Patterns

Vein-based biometrics are widely deployed in various security-based access control systems, considering their accuracy and usability. Further, vein biometrics are captured in a contactless fashion that can further enable the inflection free user verification that is highly suitable for real-life scenario. In this work, we present a novel framework based on the fusion of deep features with the pre-trained AlexNet as the backbone for reliable wristvein-based biometric verification. The proposed method is based on the AlexNet and performs the feature extraction from three different fully connected layers. Extracted deep features are further compared using the Linear Support Vector Machine (L-SVM) and combined to make the final verification decision. Extensive experiments are carried out on the dataset comprised of 100 unique wristvein patterns corresponding to 50 unique data subjects. Extensive experiments are performed with three different evaluation protocols, and the verification performance of the proposed method is benchmarked with five different deep features-based wristvein verification algorithms. The obtained results indicate the improved performance of the proposed method on all three evaluation protocols.