Exploring advantages and drawbacks of personal authentication by dorsal hand vein using hand-crafted and fast convolutional neural networks

Dorsal hand vein (DHV) recognition has emerged as a robust biometric modality due to its intrinsic resistance to forgery and high user acceptability. However, achieving high-precision authentication in dynamic environments remains a significant challenge. This study proposes an innovative, end-to-end authentication framework that integrates sophisticated preprocessing with a deep learning-based detection architecture. Our methodology utilizes an optimized image enhancement pipeline, followed by a ResNet۵۰ backbone for discriminative feature extraction, and a Faster R-CNN detector for precise vascular pattern classification. To enhance model generalization, extensive data augmentation was applied, and hyperparameters were rigorously tuned using the Nesterov Accelerated Gradient (NAG) optimizer. The framework was evaluated on the public DHV dataset, achieving state-of-the-art results with a mean Average Precision (mAP) of ۹۹.۹۸۶% and precision/recall metrics exceeding ۹۹.۹۸%. Comparative analysis against prominent models—including YOLOv۵, SSD, RetinaNet, and EfficientDet—demonstrates that our proposed approach provides superior stability, faster convergence, and exceptional accuracy in localized pattern recognition. These findings validate the framework’s reliability for high-security biometric deployments and demonstrate its potential to advance the state-of-the-art in robust personal authentication.

کلیدواژه ها:

Authentication ، biometrics ، dorsal hand vein ، deep learning ، Faster R-CNN