An Electrooculography Analysis in the Time-Frequency Domain Using Morphological Component Analysis Toward the Development of Mobile BCI Systems

摘要

Morphological Component Analysis (MCA) extended the traditional concept of signal decomposition and reconstruction by using "basis." The use of a basis not only guarantees accuracy in the reconstruction process but also requires the uniqueness of the representation using the basis. By admitting a redundancy in representations of a signal i.e. as a way of decomposition, MCA introduced the concept of a "dictionary", which includes mixtures of traditional basis. This method is frequently applied to biological signals and natural and complex image processing. In the present study, we applied MCA to decompose real electrooculography (EOG) in the time-frequency domain, which includes the electroencephalogram (EEG) signal, noise originating from measurement tools and cables for signal transmission and amplification, power-supply instability, biological fluctuations and so on. In our analysis using MCA, the EOG was decomposed into separate signal sources that could be represented using a linear expansion of waveforms from redundant dictionaries: DIRAC, UDWT and DCT. MCA was performed over several iterations to reduce the error in reconstruction. During this process; decomposed signals exhibited different characteristics in the time-frequency domain. By stopping the iteration when the correlation coefficient between the original and reconstructed signals reached a maximum (0.989 as the average), the DIRAC, UDWT and DCT represent irregular spikes, smooth curve in both the frequency and time domains and high-pass filtered components, respectively. Our results demonstrate successful decomposition via MCA and, consequently, authenticate it as an effective tool for the removal of artifacts from raw EOG signals.