User identification with memory-related brain signals

Abstract

The efforts for using biometric signature for human authentication have been long lasted and increased. Nowadays, some of the human biometric features are being used as codes for authentic systems. However, those codes still have chances to be stolen and duplicated. As the most powerful biometric signature, we suggest brain activities with memories contained in human brain. They are significantly complex, hard to be manipulated. In this research, visual stimuli were presented to evoke electroencephalographic (EEG) signals, which contain subjects’ memory information. It was expected that each of the subject would show different responses for same stimuli because of the memory differences. Features were extracted from each response with two methods

averaging signals of certain time range and discrete cosine transform (DCT). Firstly, we tried to separate the responses into the two classes

responses from known or familiar stimuli and those from unknown or unfamiliar stimuli. Support vector machine (SVM) classifier was used for the discrimination. Responses from each subject were classified separately and most of the case showed that the responses can be discriminated. Finally, the system identified the subjects by observing correlation coefficients with each of the features. As a result, up to 58 among 70 sessions were identified with right subjects. The importance of this paper lies in showing the potential of brain activities as biometric signature. In this paper, the basic methods and ideas for single-trial analysis are well described. Proposed experimental paradigm, preprocessing methods, analyzing methods and performance measure can show the direction for the further research related to brain computer interface (BCI).