Person Verification Based on Spectral Fusion of Intra-Palm Propagation Signal

We have been pursuing the feasibility of person verification using intra-palm propagation signals. When a small current is passed between electrodes on the palm of the hand, a leakage electric field is generated around them, which propagates to a distant electrode and is detected as a change in potential. This is called an intra-palm propagation signal. If the way the signal propagates varies across individuals, it can be used for biometric identification. Amplitude and phase spectra were obtained from the intra-palm propagation signal, and after smoothing and normalization, frequency components with a variance ratio of 1 or greater were extracted as individual features. Each was verified using a machine-learning one-to-one classifier, and finally, the amplitude and phase were combined to make the final decision. For the phase spectrum, we proposed a second-order differential phase spectrum, which does not require a reference signal. However, the verification error was at best 10%, and it was not possible to achieve 0%. In this study, the normalization method is revised. In addition, as an individual feature, frequency components are sorted in descending order of variance ratio. Various machine-learning-based classifiers are evaluated, and those with the highest performance are selected. One of these is then adopted as the classifier, and the amplitude and phase are integrated at the feature, score, and outcome levels. When evaluated on a newly created dataset of 33 subjects, a verification error rate of 0.087% is achieved with score-level fusion with adjusted weights, and 0.10% with result-level fusion based on OR operation.