[Palm vein recognition based on end-to-end convolutional neural network].

Area of Science:

• Biometrics
• Computer Vision
• Machine Learning

Background:

• Accurate and efficient biometric identification is crucial for security and clinical applications.
• Traditional palm vein recognition methods face challenges in accuracy and speed.
• Deep learning offers potential for advanced feature extraction and classification in biometrics.

Purpose of the Study:

• To propose a novel palm-vein recognition model utilizing an end-to-end convolutional neural network (CNN).
• To enhance the accuracy and efficiency of palm vein identification.
• To provide a new approach for palm vein recognition in clinical settings.

Main Methods:

• Developed an end-to-end CNN model with alternating convolutional and pooling layers for feature extraction.
• Employed mini-batch stochastic gradient descent with momentum for optimizing feature descriptors.
• Implemented data augmentation, batch normalization, dropout, and L2 regularization to minimize generalization error.

Main Results:

• Achieved identification rates of 99.90% on the PolyU database (500 subjects) and 98.05% on a self-established database.
• Demonstrated single-sample identification time of less than 9 milliseconds.
• The proposed CNN model significantly improved accuracy compared to traditional methods.

Conclusions:

• The novel end-to-end CNN model offers a highly accurate and efficient solution for palm vein recognition.
• The model's performance suggests significant potential for improving security and clinical applications.
• This research presents a new, effective approach to palm vein recognition using deep learning.