An EOG Artifacts Correction Method Based on Subspace Independent Component Analysis

The ocular artifact correction based on Linear Combination and Regression method is widely used in electroencephalographic (EEG) processing. Its main idea is to subtract a fraction of the recorded electro oculography (EOG) from the EEG signal to remove the overlapped EOG noise from EEG. But traditional regression method ignored the EEG signals mixed in the recorded EOG, and thus the overcorrection process will lead to some EEG signal lost during the correction of EOG. This paper proposed a new method of EOG correction based on Subspace Independent Components Analysis (SICA) which considers the Bi-directionality between the EOG and the EEG signal. Here, the advanced model is used to reduce the EOG artifacts from the datasets of the simulate data and the EEG dataset from BCI competition IV. With the different evaluation standards and performance index, the commonly used regression methods and the modified method were compared, and results indicated that EOG is well removed from EEG signals by our modified methods.