Off-the-person ECG Biometrics Using Convolutional Neural Networks

Off-Person ECG Biometrics Using Spatial Representations and Convolutional Neural Networks

“…The paper extends the preliminary results published in [29], by including 3 additional spatial representation methods (S-Transform, Gramian Angular Field, recurrence plot) besides the state-space portraits approach (while the latter now additionally considers 3D state-space trajectories, as opposed to 2D input images in [29]), analyzing verification EER performances besides identification accuracy, and studying the effect of input data fusion and long-term test data.…”

“…Much similar to [37], each time-delayed version of the (filtered, R-peak centered, normalized, fixed-length) ECG segment is first scaled into a common range, then a 200x200 pixels binary image is obtained by setting black color on all pixels hit by at least one point of the phase space trajectory, while labeling by white color all the rest of the pixels. As a consequence, the subsequent CNN based classifier may use single channel bidimensional inputs (as analyzed in [29]) or three (generally, multiple) channel inputs obtained by projecting the 3D (generally multidimensional) representation onto corresponding orthogonal planes.…”

Convolutional Neural Network for Individual Identification Using Phase Space Reconstruction of Electrocardiogram

“…PSR coverts morphological characteristics (temporal patterns) in an ECG signal into phase-space loops (spatial patterns). The converted spatial patterns are different among individuals, leading to good performances in ECG biometrics [ 12 , 13 , 22 , 25 ]. In particular, each spatial pattern has specific locations in the phase space, allowing 2D CNN to efficiently capture the phase-space fingerprints.…”

“…A simple method is to put single-beat ECG samples in a vector, then one-dimensional (1D) convolutional networks are used to produce distinguishing features for a succeeding connected network to identify individuals [ 19 ]. The other methods transform the 1D ECG signals to two-dimensional (2D) images by plotting each ECG beat as an individual grayscale image (time–amplitude representation) [ 20 ], decomposing each ECG beat on various scales using continuous wavelet transform (time–scale representation) [ 21 ], and embedding each ECG beat onto a 2D phase-space image using a time-delay technique [ 22 ]. In addition, multi-beat ECG is also transformed to an image by stacking beat-aligned amplitudes [ 23 , 24 ] or merging multiple phase-space trajectories [ 25 ].…”

“…Thus, several studies employed the PSR with m = 2 to detect QRS complex [ 28 ], recognize ventricular extrasystoles [ 29 ], and classify the type of ventricular arrhythmia [ 30 ] because the 2D reconstruction displays a concise phase space trajectory and lends itself more readily to feature extraction. In addition, PSR transforms the temporal patterns of the ECG signal (P, QRS, and T waves) to specific spatial loops in phase space without the need of characteristic wave detections, extending its application to individual identification [ 12 , 13 , 22 , 24 ].…”