Scholarlink

<p>Biometric authentication systems have become a vital component of modern security infrastructures due to their reliability in personal identification. However, traditional unimodal biometric systems often suffer from high error rates, noise sensitivity, and vulnerability to spoofing attacks. To address these shortcomings, this research suggests a Hybrid Ensemble Framework of Multimodal Biometric Authentication, which combines several traits including face, iris, and palm to provide a greater accuracy and strong range of endurance. The Local Binary Patterns (LBP) is utilized as the methodology used to extract the texture-based feature, and VGG-16 and the Vision Transformer (ViT) are used in conjunction to extract both global and local context features. The model is trained and validated using the MULB dataset, which has 11,280 images of 188 people. The fusion strategy at feature level and score level are used to achieve the best performance with respect to modalities. The experimental outcomes reveal high recognition accuracy of 99.12, precision of 99.30, recall of 99.05 and F1-score of 99.17 that are much high than the traditional CNN and uni-modal models. The suggested framework has a high reliability, robustness and security of adversarial biometric conditions.</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

Hybrid ensemble framework for multimodal biometric system using binary patterns and deep learning

  • 期刊论文

  • 发表日期：

    2025年12月27日

查看全文
  • Laxman Singh,
  • Ashish Kumar,
  • Richa Golash

摘要

Biometric authentication systems have become a vital component of modern security infrastructures due to their reliability in personal identification. However, traditional unimodal biometric systems often suffer from high error rates, noise sensitivity, and vulnerability to spoofing attacks. To address these shortcomings, this research suggests a Hybrid Ensemble Framework of Multimodal Biometric Authentication, which combines several traits including face, iris, and palm to provide a greater accuracy and strong range of endurance. The Local Binary Patterns (LBP) is utilized as the methodology used to extract the texture-based feature, and VGG-16 and the Vision Transformer (ViT) are used in conjunction to extract both global and local context features. The model is trained and validated using the MULB dataset, which has 11,280 images of 188 people. The fusion strategy at feature level and score level are used to achieve the best performance with respect to modalities. The experimental outcomes reveal high recognition accuracy of 99.12, precision of 99.30, recall of 99.05 and F1-score of 99.17 that are much high than the traditional CNN and uni-modal models. The suggested framework has a high reliability, robustness and security of adversarial biometric conditions.