ECG and EEG Based Multimodal Biometrics for Human Identification

Developing multi-biometric systems using multi-modal signals is the recent trend in biometric identification problem. Integrating heart and brain electrical signals (ECG and EEG) is very important because of their liveliness property and robustness against falsification. In this study, we have investigated the fusion of ECG and EEG signals from low-cost devices with multiple classifiers (KNN, LDA, and ESAVM) using wavelet domain statistical feature. After preprocessing, multiscale wavelet packet decomposition is applied to the signal (ECG/EEG) segment. Feature vectors are computed from the transformed signal using statistical descriptors, called wavelet packet statistics (WPS). ECG and EEG traits are fused at feature level, while two classifiers are fused at the decision level. An experiment with ten human subjects showed promising results of human identification using fused trait (ECG-EEG) with fused classifiers. The fused trait i.e., the fused WPS vectors from ECG and EEG signals with the fused classifier produces the highest average Fscore (90.5%), when compared with the single trait (ECG or EEG) with single classifier (54.7% with ECG; 74.9% with EEG) or single trait with fused classifier (66.0% with ECG; 87.3% with EEG). A brief ROC analysis also confirmed the above findings.