Ocular Artifact Removal Based on Local Mean Decomposition and Canonical Correlation Analysis

Based on local mean decomposition (LMD) and canonical correlation analysis (CCA), an automatic removal method, denoted as LMDC, was proposed to eliminate the serious impact of ocular artifact (OA) from electroencephalography (EEG). Each recorded EEG was decomposed into a series of physically meaningful production function (PF) components adaptively by LMD, and CCA was applied to eliminate the correlation among the PFs to get the corresponding canonical correlation variable. Then, the correlation coefficient matrix between each EEG and multi electrooculogram (EOG) was computed to recognize the OA component automatically. The random variables corresponding with OA components in the canonical correlation variable were set to zero, and the others remain unchanged to obtain a new canonical correlation variable. Finally, the inverse algorithm of CCA was utilized to project the new canonical correlation variable to the OA free PFs, and the OA removed EEG was reconstructed. Experimental research was conducted on a public brain computer interface (BCI) completion database. Experiment results show that LMDC has better performance than that of the other related methods, and has stronger adaptability for multi subjects and types of OA.