Estimation of EOG artifacts from EEG signal by Orthogonal Matching Pursuit and thresholding

Electroencephalogram (EEG) signals are consistently contaminated by Electrooculogram (EOG) signals or by unavoidable eye-blink artifacts. Efficient separation of eye-blink artifacts from EEG signal is challenging for mostly two reasons-1) EOG and EEG signals are spectrally overlapped in low frequency zone, 2) uncontaminated regions of EEG signal should remain unaltered after EOG separation for neural leakage suppression. To overcome these two challenges simultaneously, in this work we have estimated the EOG artifacts from contaminated EEG signal, through employing sparse decomposition by Orthogonal Matching Pursuit [OMP] and thresholding of corresponding coefficients. Sine and cosine waveforms of densely spaced different frequencies are utilized as dictionary atoms. OMP representation of small overlapped contaminated EEG segments with relaxed sparsity bound, exhibit a distinct pattern in concatenated coefficient vector. This specific pattern leads to a straightforward thresholding, for unnecessary coefficient rejection. Time and frequency domain results along with performance measurements, authenticate the effectiveness of proposed method.