Multifractal feature vectors for Brain-Computer interfaces

This article introduces a new feature vector extraction for EEG signals using multifractal analysis. The validity of the approach is asserted on real data sets from the BCI competitions II and III. The feature extraction can be performed in real time with low-cost discrete wavelet transforms. Classification results obtained with the new feature vectors are close to the state of art techniques, while using a different information. Combining the new multifractal feature vector with existing ones may result in better performances, up to 5% in the present case. This work thus offers an alternative to the usual feature-extraction techniques, and opens new possibilities in the field of Brain-Computer interfaces.