Dataset of 2D ECG Features (Binary Images and STFT Spectrograms) for Biometric Identification

This dataset provides the complete set of 2D features and preprocessed signals used in the scientific publication titled "ECG in security practice: Biometric identification of individuals based on heart signals". The data is designed to support reproducible research in ECG-based biometrics, deep learning, and signal processing.

The dataset is derived from two publicly available sources:

ECG-ID Database (PhysioNet)

Heartprint Database

The raw signals from these sources were subjected to a rigorous preprocessing pipeline, which included:

Signal Conditioning: Third-order Butterworth band-pass filtering to remove noise and baseline wander correction using a 16th-order polynomial fit.

Segmentation: R-peak detection using the Pan-Tompkins algorithm, followed by R-centered segmentation to extract individual heartbeats.

Quality Control: Manual and algorithmic removal of corrupted or noisy beats.

From these cleaned and segmented heartbeats, two distinct types of 2D features were generated for training convolutional neural networks (CNNs):

Binary Images: Morphology-preserving binary images (512x256 px) that directly represent the time-domain waveform shape.

STFT Spectrograms: Time-frequency representations (800x600 px) generated using a Short-Time Fourier Transform with a Hann window.

The repository is structured into folders containing the original signals, the segmented beats, and the final 2D feature sets for both the ECG-ID and Heartprint databases. A supplemental folder, Spectrograms_Test, contains spectrograms generated with varying parameters for robustness analysis.

A detailed metadata file, description_final_database_of_input_images.xlsx, is also included, providing per-subject statistics, including the number of available recordings and segmented beats, which is essential for replicating the experimental setup.