ECG Biometrics via Dual-Level Features with Collaborative Embedding and Dimensional Attention Weight Learning.

Area of Science:

• Biometrics
• Signal Processing
• Machine Learning

Background:

• Electrocardiogram (ECG) biometrics is gaining traction for individual identification.
• Current methods often rely solely on 1D time-series features, limiting recognition accuracy.
• Enhanced feature extraction is crucial for optimal ECG biometric performance.

Purpose of the Study:

• To propose a novel framework for ECG biometric recognition by integrating dual-level features (1D and 2D).
• To improve the discriminability of individual identification in ECG biometrics.
• To develop an effective optimization algorithm for the proposed framework.

Main Methods:

• Integration of 1D (time series) and 2D (relative position matrix) ECG representations.
• Utilizing collaborative embedding, dimensional attention weight learning, and projection matrix learning.
• Employing collective matrix factorization for shared latent representation learning.

Main Results:

• The proposed method effectively integrates dual-level features, enhancing representation discriminability.
• Dimensional attention learning preserves diverse information across latent representation dimensions.
• Experimental results on benchmark datasets demonstrate superior performance compared to state-of-the-art methods.

Conclusions:

• The novel dual-level feature integration framework significantly improves ECG biometric recognition.
• The method offers enhanced accuracy and discriminability for individual identification.
• The developed optimization algorithm is effective and efficient for practical applications.