EEG-Based Biometric Recognition Based on Wavelet Packet Analysis and Manifold Learning

Compared to traditional biometric identification methods, EEG-based person recognition typically exhibits lower accuracy. In this paper, we present the study to utilize manifold learning for biometric identification based on resting-state EEG signals. We employ the Daubechies wavelet packet transform to decompose the original EEG signals into four rhythms, and subsequently utilize fast Fourier transform to extract the signal energy spectrum feature vector. Thereafter, manifold learning is applied to reduce the dimensionality of the extracted features, with various machine learning algorithms employed as classifiers. Furthermore, we analyze the impact of different combinations of machine learning methods and manifold learning settings. Our methodology is validated using a 109-subject EEG Motor Movement/Imagery dataset, achieving an accuracy rate of 98.2%.