ECG Biometrics on Mobile Devices: High-Accuracy Authentication Using i-Vectors and Cepstral Coefficients

Abstract

In recent years, the growing importance of personal data security has led to a rapid increase in demand for biometric authentication systems. This study proposes a biometric authentication method based on electrocardiogram (ECG) data that holds potential for use in mobile devices. The proposed approach integrates Mel Frequency Cepstral Coefficients (MFCC) and Gammatone Frequency Cepstral Coefficients (GFCC) features with the i-vector technique to effectively capture the biometric information present in the ECG signal within the 1-35 Hz range. In particular, the complementary effect of GFCC at low frequencies enhances the robustness of the MFCC-based features, thereby providing a more reliable biometric representation. The method was evaluated using the ECG-ID, Heartprint, and a dataset collected specifically for this study. While all datasets were obtained at different time intervals, the Heartprint dataset also includes recordings from multiple sessions. Performance metrics obtained from the Heartprint dataset-namely, 94.39% accuracy, 93.17% true acceptance rate (TAR), and 5.61% false acceptance rate (FAR)-demonstrate that the i-vector-based approach yields results on large datasets that are comparable to those achieved by deep learning and conventional machine learning methods. The use of a brief 5-second ECG signal minimizes memory and processing power requirements, enabling rapid data processing and real-time authentication. The data collected specifically for this study were acquired using the cardiochip BMD101; this sensor offers easy integration with wearable devices such as smartwatches or phone cases, thereby supporting the feasibility of applying the proposed method on mobile platforms. In conclusion, the integration of MFCC and GFCC features with the i-vector technique provides an effective and mobile-compatible solution for ECG-based biometric authentication, while also establishing a solid foundation for future developments in biometric systems.