Analysis and classification of visually evoked EEG using multiresolution approximation

Electroencephalogram (EEG) is the electrical activity of the brain acquired by recording from electrodes placed on the scalp. The EEG has many usages: It is used to construct a communication channel between human being and his environment; it helps to understand the brain functions; it is used in diagnosis and therapy. Visual Evoked potentials (VEPs) in the EEG that occurs in response to a visual stimulus are a tool to facilitate such advancements.The objective of this work is to analyze and classify VEPs obtained from familiar and unfamiliar face experiments. EEG signals were recorded from 26 volunteers. Multi-resolution analysis was used as a tool for signal approximation and time-frequency analysis. A custom scaling-wavelet function pair and its bi-orthogonal complements were built by resembling the waveform of the scaling function to excitatory post-synaptic potential (EPSP). The approximation coefficients of the VEPs were obtained from the custom scaling function, and the approximation coefficients of the training set were fed to a Fisher's linear classifier to distinguish the familiar-unfamiliar face specific VEP. The classification performance of the proposed wavelet has been slightly higher than the well-known wavelets with filter lengths which are close to the filter length of the custom scaling function. Furthermore, some EEG data features and common spatial patterns methods were also investigated. The performances of the custom multi-resolution system, ordinary wavelets, the EEG features and, the common spatial patterns methods were examined and compared. The results show that the classification performance of proposed technique has promising success for the EEG signal representation and classification.