Introduction of Fractal Dimension Feature and Reduction of Calculation Amount in Person Authentication Using Evoked EEG by Ultrasound

Abstract

The aim of this study is to authenticate individuals using an electroencephalogram (EEG) evoked by a stimulus. EEGs are highly confidential and enable continuous authentication during the use of or access to the given information or service. However, perceivable stimulation distracts the users from the activity they are carrying out while using the service. Therefore, ultrasound stimuli were chosen for EEG evocation. In our previous study, an Equal Error Rate (EER) of 0 % was achieved; however, there were some features which had not been evaluated. In this paper, we introduce a new type of feature, namely fractal dimension, as a nonlinear feature, and evaluate its verification performance on its own and in combination with other conventional features. As a result, an EER of 0 % was achieved when using five features and 14 electrodes, which accounted for 70 support vector machine (SVM) models. However, the construction of the 70 SVM models required extensive calculations. Thus, we reduced the number of SVM models to 24 while maintaining an EER = 0 %.