Feature fusion and classification of EEG/EOG signals

Abstract

Electroencephalogram (EEG) refers to the brain waves, whereas electrooculogram (EOG) represents the eyeblinking signals. Both the signals possess complexities and various artifacts when they are recorded. In order to use these signals in biometric and clinical applications, preprocessing needs to be done. Stationary wavelet transform (SWT) with the combination of independent component analysis (SWT + ICA) is used to perform EEG signal preprocessing, and empirical mode decomposition (EMD) is used to process EOG data. After the signals are alleviated of the noise, feature extraction is done. Time delineation in case of EOG and autoregressive modeling (AR) technique in case of EEG data is applied for feature extraction. Fusion of extracted features is performed using canonical correlation analysis (CCA) so that the number of features is minimized. Classification is performed to classify the features into sets or classes in order to perform dimensionality reduction. Linear discriminant analysis (LDA) is used to form suitable sets and evaluates the performance of the classifier. EOG signal is present as an artifact in EEG; but in this paper, their combination is considered so as to gain more distinct features. The main aim is to develop a multimodal system which possesses high classification and recognition accuracy so that biometric authentication can be performed.