Feature selection and channel optimization for biometric identification based on visual evoked potentials

In recent years, biometric identification has received general concerns around the world, and become a frontal and hot topic in the information age. Among the internal biometric traits, electroencephalogram (EEG) signals have emerged as a prominent characteristic due to the high security, uniqueness and impossibility to steal or mimic. In this paper, individual difference of visual evoked potentials (VEPs) with cognition task were investigated, in addition, a feature selection and channel optimization strategy was developed for the VEPs based biometric identification system, where three different methods, including genetic algorithm (GA), Fisher discriminant ratio (FDR), and recursive feature elimination (RFE) were employed. In our experiments with 20 healthy subjects, the classification accuracy by support vector machine (SVM) reached up to 97.25% with AR model parameters, compared to 96.25% before optimization, and 32 channels of most discriminative were eventually selected from 64 channels with best performance. Results in this study revealed the feasibility of VEPs based EEG to be used for biometric identification. The proposed optimization algorithm was shown to have the ability to effectively improve the identification accuracy as well as simplifying the system. Further investigate may provide a novel idea for the individual difference analysis of EEG and for its practical design and optimization in the field of biometrics in the future.