Abstract:
A Biometric system utilizes biological characteristics like DNA, Voice, Ear, Face, Finger prints to uniquely establish user identity. Brain Computer Interface (BCI) based Biometric system relies on unique neural structure and mental activity signals commonly known as Electroencephalogram (EEG) signals to authenticate the user and are inherently immune to impersonation.
Details learned from existing applications of Fractional Fourier Transform (FRFT) were used to devise an algorithm that utilized fractional coefficients extracted from subject specific EEG patterns for various tasks performed by the user. The advantage of using Fractional Fourier Transform is that it offers improved performance for systems that are based on Fourier Transform with little additional cost and as it’s a linear one dimensional Time-Frequency distribution it is more suitable for time-varying EEG signals than Fourier Transform. Exact Radial Basis (RBE) Neural Network was used for classification due to its advantage of low training time. A special one dimensional case of k-means clustering was used to calculate the threshold in order to accept the user as a client or reject him as an imposter. The proposed system works on signals recorded from multiple channels, therefore Weight Adjustment was carried out for each electrode and task performed by the user.
The performance improvements as a result of weight adjustment and the efficiency of the proposed algorithm were tested on pre-recorded dataset available from Colorado University and another dataset was self-recorded using the indigenously designed hardware prototype. The average accuracy achieved for both the datasets was more than 75% with response time of less than two seconds.
