A Method of Designing Sparse Dictionary for Event-related Potential

Sparse decomposition is a trending and reliable paradigm for EEG analysis. Among the common algorithms for sparse decomposition and representation, K-SVD is the most popular one. However, K-SVD cannot extract complete event-related component from EEG into dictionary atoms, and components are usually distributed among many atoms. In this paper we propose a method that combine K-SVD with Sparse Performance Index as constraint, to solve the problem that dictionary atoms are not related to ERP components. Then we propose procedures to analysis EEG data with proposed method. Finally, we demonstrated the capabilities of this method by applying it to a public EEG dataset that contains auditory evoked potential. The result showed that our method can extract complete ERP waveforms into atoms.