Scholarlink

<p>Biometric authentication, a pivotal means of identity verification, has witnessed significant interest, offering advantages over conventional password-based systems. Electrocardiogram (ECG) biometrics, an emerging authentication technique, harnesses the heart’s electrical signals for robust identity validation. To address privacy and security concerns associated with ECG data, we propose the Cancelable Privacy-Preserving Biometric Obfuscation (CPBO) method integrated with deep learning for ECG authentication. By using cryptographic hash functions along with randomization based on Principal Component Analysis (PCA), CPBO converts ECG templates into <i>unlinkable</i> versions. This approach protects user privacy while preserving the system’s ability to authenticate identities effectively. The novelty of our work lies in the holistic integration of CPBO with a tailored deep learning model within a unified framework, creating a transparent and secure system that jointly optimizes feature learning and template protection. This integration improves the overall performance of the authentication system while effectively addressing issues related to interpretability. Although cancelability and privacy are often considered related, this study treats them as distinct yet complementary concepts. Cancelability allows for the revocation and replacement of compromised templates, whereas privacy preservation focuses on protecting sensitive ECG data from misuse or identity tracking. Experimental evaluations conducted on three publicly available ECG databases demonstrate strong results across key performance metrics, including accuracy, precision, recall, F1-score, and equal error rate (EER). The proposed framework satisfies all cancelable biometric criteria. It provides a secure and interpretable ECG authentication model. The design is lightweight, with few parameters and fast inference. This makes it ideal for real-time applications.</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

Cancelable and privacy-preserving ECG biometrics with deep learning for enhanced authentication

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Mohamed Hammad,
  • Souham Meshoul,
  • Nebojsa Bacanin,
  • Paweł Pławiak,
  • Sondos Fadl

摘要

Biometric authentication, a pivotal means of identity verification, has witnessed significant interest, offering advantages over conventional password-based systems. Electrocardiogram (ECG) biometrics, an emerging authentication technique, harnesses the heart’s electrical signals for robust identity validation. To address privacy and security concerns associated with ECG data, we propose the Cancelable Privacy-Preserving Biometric Obfuscation (CPBO) method integrated with deep learning for ECG authentication. By using cryptographic hash functions along with randomization based on Principal Component Analysis (PCA), CPBO converts ECG templates into unlinkable versions. This approach protects user privacy while preserving the system’s ability to authenticate identities effectively. The novelty of our work lies in the holistic integration of CPBO with a tailored deep learning model within a unified framework, creating a transparent and secure system that jointly optimizes feature learning and template protection. This integration improves the overall performance of the authentication system while effectively addressing issues related to interpretability. Although cancelability and privacy are often considered related, this study treats them as distinct yet complementary concepts. Cancelability allows for the revocation and replacement of compromised templates, whereas privacy preservation focuses on protecting sensitive ECG data from misuse or identity tracking. Experimental evaluations conducted on three publicly available ECG databases demonstrate strong results across key performance metrics, including accuracy, precision, recall, F1-score, and equal error rate (EER). The proposed framework satisfies all cancelable biometric criteria. It provides a secure and interpretable ECG authentication model. The design is lightweight, with few parameters and fast inference. This makes it ideal for real-time applications.