PPG-based Biometric Identification: Discovering and Identifying a New User.

Area of Science:

• Biometrics
• Signal Processing
• Machine Learning

Background:

• Photoplethysmography (PPG) signals from wearable devices are increasingly used for biometric identification.
• Current PPG biometric identification systems primarily focus on verification and struggle with identifying new users without retraining.
• Existing methods require model retraining for new users, reducing overall identification accuracy.

Purpose of the Study:

• To investigate a deep neural network approach for identifying both existing and new users using PPG signals.
• To develop a method that can discover and identify new users without the need for model retraining.
• To evaluate the performance of the proposed approach on a public dataset.

Main Methods:

• Utilized a deep neural network as a feature extractor for PPG signals.
• Employed feature vector distance to discover and identify new users.
• Trained the deep neural network exclusively on data from existing users.

Main Results:

• Achieved over 99% accuracy for identifying existing users on the BIDMC dataset.
• Demonstrated over 90% accuracy in discovering new users.
• Obtained an average accuracy of approximately 90% for identifying new users.

Conclusions:

• The proposed deep neural network approach effectively identifies existing users.
• The method shows feasibility and high accuracy in discovering and identifying new users without retraining.
• This approach offers a viable solution for robust biometric identification in wearable devices.