Scholarlink

This paper proposes that biometric template security remains a pressing concern in modern authentication systems, particularly under adversarial threats like template inversion or spoofing. This paper proposes a novel and secure biometric authentication framework that leverages modality-specific feature extraction and Adaptive BioHashing guided by a Deep Reinforcement Learning (DRL) agent. The system employs Scale-Invariant Feature Transform (SIFT) for facial features and Histogram of Oriented Gradients (HOG) for iris data. A DRL agent dynamically selects hashing parameters based on recognition performance and simulated attack risk, resulting in enhanced template robustness. Classification is performed using a Support Vector Machine (SVM). Experimental evaluations on IIT Delhi datasets show a superior accuracy of 98.20% and a low Equal Error Rate (EER) of 1.2%, outperforming existing state-of-the-art methods. The proposed framework demonstrates strong potential for secure, adaptive biometric systems.

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

Adaptive BioHashing via Reinforcement Learning: A Dynamic Template Protection Framework for Multimodal Biometrics

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • M. Prakasha,
  • K. Vannurswamy,
  • G. Hemantha Kumar

摘要

This paper proposes that biometric template security remains a pressing concern in modern authentication systems, particularly under adversarial threats like template inversion or spoofing. This paper proposes a novel and secure biometric authentication framework that leverages modality-specific feature extraction and Adaptive BioHashing guided by a Deep Reinforcement Learning (DRL) agent. The system employs Scale-Invariant Feature Transform (SIFT) for facial features and Histogram of Oriented Gradients (HOG) for iris data. A DRL agent dynamically selects hashing parameters based on recognition performance and simulated attack risk, resulting in enhanced template robustness. Classification is performed using a Support Vector Machine (SVM). Experimental evaluations on IIT Delhi datasets show a superior accuracy of 98.20% and a low Equal Error Rate (EER) of 1.2%, outperforming existing state-of-the-art methods. The proposed framework demonstrates strong potential for secure, adaptive biometric systems.