ECG Biometric Verification with Machine-Learning Noise Filtering

The patent applies preliminary action by performing baseline wander removal and discrete wavelet transforms on ECG signals before authentication occurs. These preprocessing steps are executed in advance to eliminate noise and stabilize the signal, ensuring that the subsequent authentication process operates on clean, reliable data rather than raw, noisy signals.

The patent uses an intermediary approach by introducing a machine learning classifier as a mediator between the raw ECG signal and the authentication decision. The classifier processes the preprocessed signal features and determines whether the signal corresponds to a legitimate user, effectively filtering out false positives caused by noise and cardiac variations before authentication is granted.

The patent applies dynamics by using a machine learning classifier that adapts to the variable nature of ECG signals. Rather than relying on static template matching that fails when signals vary, the classifier dynamically learns patterns from training data and can recognize legitimate users even when their ECG signals differ due to cardiac variations, thereby maintaining both security and precision.

The patent implements feedback by using the training set to continuously refine the machine learning model's understanding of legitimate ECG patterns. The system learns from the feedback provided by training data about what constitutes normal cardiac variation, allowing it to distinguish between legitimate variability and fraudulent attempts, thus maintaining high matching precision despite dynamic signal characteristics.