A new method to separate two components, background and spiky activity, from subdural electroencephalogram (EEG) is presented in this paper. The spiky activity is characterized by transient waveforms whose spectra are overlapped with the background activity which is the dominant activity with relatively slow amplitude. Both components are non-stationary. It is found that the two components are different in their morphological characteristics. Based on this difference we apply nonlinear morphological operations to scaling and wavelet coefficients so as to obtain an accurate, separated reconstruction of the two components. This is based on repartitioning and manipulating energy in the localized scaling and wavelet coefficients so that key characteristics of the spiky activity appear in one sub-signal while characteristics of the background activity appears in another sub-signal. By using a simple morphological operation on both scaling coefficients and wavelet coefficients, our preliminary investigation produced promising results.
