Scholarlink

<p>Biometric-based personal authentication is actively being researched in various fields that require high reliability and security. Currently, research on vein patterns for biometric recognition has primarily focused on the veins in the palm or fingers. However, the hand area is highly exposed to external factors, making it susceptible to counterfeiting, and it has the drawback that biometric information can easily be altered due to injuries or skin damage. To overcome these limitations, this study proposes a new biometric approach utilizing vein patterns in the neck, which is relatively less exposed, maintains a stable form, and is less prone to changes. The objective of this study is to compare and evaluate the performance of deep learning and machine learning models for neck vein pattern recognition by combining them with preprocessing techniques. The neck vein pattern images utilized in this experiment underwent preprocessing steps, including region of interest (ROI) extraction, contrast enhancement using CLAHE (Contrast Limited Adaptive Histogram Equalization), vessel structure enhancement through the Frangi filter, and noise reduction via the Bilateral filter. The preprocessed dataset was then trained and evaluated using four models: CNN, SVM, EfficientNet-B7, and Vision Transformer. The results showed that the SVM model achieved 94.82% accuracy, the CNN model achieved 96.13%, the EfficientNet-B7 model achieved 97.50%, and the Vision Transformer model achieved 98.20%, with the Vision Transformer model demonstrating the best performance in neck vein pattern recognition. These findings not only validate the feasibility of neck vein pattern-based biometric technologies but also offer significant benchmarks for future advancements and broader adoption of personal authentication systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Performance Comparison of Deep Learning Models Based on Neck Vein Patterns for Biometric

  • 期刊论文

  • 发表日期：

    2026年3月23日

查看全文
  • Rae-Hyun Yu,
  • Kyung-Ho Kim

摘要

Biometric-based personal authentication is actively being researched in various fields that require high reliability and security. Currently, research on vein patterns for biometric recognition has primarily focused on the veins in the palm or fingers. However, the hand area is highly exposed to external factors, making it susceptible to counterfeiting, and it has the drawback that biometric information can easily be altered due to injuries or skin damage. To overcome these limitations, this study proposes a new biometric approach utilizing vein patterns in the neck, which is relatively less exposed, maintains a stable form, and is less prone to changes. The objective of this study is to compare and evaluate the performance of deep learning and machine learning models for neck vein pattern recognition by combining them with preprocessing techniques. The neck vein pattern images utilized in this experiment underwent preprocessing steps, including region of interest (ROI) extraction, contrast enhancement using CLAHE (Contrast Limited Adaptive Histogram Equalization), vessel structure enhancement through the Frangi filter, and noise reduction via the Bilateral filter. The preprocessed dataset was then trained and evaluated using four models: CNN, SVM, EfficientNet-B7, and Vision Transformer. The results showed that the SVM model achieved 94.82% accuracy, the CNN model achieved 96.13%, the EfficientNet-B7 model achieved 97.50%, and the Vision Transformer model achieved 98.20%, with the Vision Transformer model demonstrating the best performance in neck vein pattern recognition. These findings not only validate the feasibility of neck vein pattern-based biometric technologies but also offer significant benchmarks for future advancements and broader adoption of personal authentication systems.