An Enhanced Finger Vein and Palm Vein Authentication System Based on a Hybrid PCANet with Inception-ResNet Models

Abstract

Finger vein and palm vein authentication are gaining significance for their security, ease of use, and superior accuracy. Subcutaneous vein patterns are more secure than iris, fingerprints, retina, and facial patterns and cannot be duplicated or forged. Advances in deep learning have enhanced speed and accuracy, making vein-based authentication a secure access system alternative. PCANet deep learning is used in finger vein and palm vein authentication to extract deep features with low computational cost, while Inception-ResNet improves deep feature representation with its residual connections and inception models. This integration improves the accuracy, strength, and effectiveness of secure biometric authentication by combining both models. This paper proposes an enhanced finger vein and palm vein authentication system (EFVPV-AS) that depends on two hybrid deep learning and hyperparameter machine learning models. To improve image quality, the proposed EFVPV-AS undergoes three methods. Second, EFVPV-AS adopts efficient and light-weight hybrid deep learning models: unsupervised PCANet with supervised Inception-ResNet-V1 and V2 to extract unique features from each modality by considering the finger and palm veins. It uses all features extracted from these models to concatenate a fusion authentication of the finger and palm veins. Finally, a pair of machine learning authentication models: model parameter and hyperparameter in conjunction with KNN, SVM, and RF. The EFVPV-AS experiments initially used PLUSVein-Contactless Finger and Hand Veins RL850nm and RL950nm datasets. The system achieved the best performance on this dataset, with an F1_Score of some metrics of 99.82% for KNN, 100% for SVM, and 100% for RF. Furthermore, to verify its generalizability, additional experiments were conducted on a standard public dataset (University of Twente Finger Vascular Pattern (UTFVP), and CASIA multi-spectral palm print v1.0), where the system showed consistently high performance with an F1_Score (95.34%, 96.31% for KNN, 99.82%, 100% for SVM, and 91.15%, 98.76% for RF). These results cover the hyperparameter model, which concatenates the fusion of all features extracted from the finger and palm vein. With superior image quality, strong vein-specific features, and matching parameters, the EFVPV-AS outperforms the most advanced finger and palm vein authentication systems.