Untouchable and Cancelable Biometrics: Human Identification in Various Physiological States Using Radar-Based Heart...

Area of Science:

• Biometrics
• Signal Processing
• Healthcare Security

Background:

• Biometric data in healthcare face privacy and security risks during network transmission.
• Wearable devices, smartphones, and IoT increase data vulnerability.
• Cancelable biometrics offer a solution for secure, privacy-preserving identification.

Purpose of the Study:

• To propose a cancelable identification model using contactless heart signals.
• To evaluate the model's accuracy and privacy preservation capabilities.
• To compare different feature extraction strategies for improved performance.

Main Methods:

• Acquisition of contactless heart signals using continuous-wave radar.
• Signal transformation into scalograms and feature extraction using Convolutional Neural Networks (CNNs).
• Creation of cancelable templates via Gaussian Random Projection (GRP) and classification using Multilayer Perceptron (MLP).

Main Results:

• Achieved 91.20% accuracy across all scenarios, increasing to 95.40% in resting scenarios.
• CNNs trained exclusively on the dataset outperformed transfer learning models in feature extraction.
• Demonstrated the feasibility of the proposed cancelable biometric system.

Conclusions:

• The proposed cancelable biometric system effectively secures heart signal data.
• Contactless radar-based heart signals combined with CNNs offer a robust identification method.
• The system enhances privacy and security in healthcare biometric applications.