Abstract
In this paper, we introduce new features that quantitatively characterize the shape of the m-dimensional phase space trajectory reconstructed for the electroencephalogram (EEG) signals which reflects the brain electrical activity at different motor imagery tasks. The proposed features consist of the distances between the two extreme points along each embedding dimension of the reconstructed phase space (RPS) as well as the length of the line segment representing the projection of the trajectory points on every embedding dimension separately. The new features were extracted for dataset III from BCI competition II while the K-nearest neighbor (KNN) classifier was used to evaluate the effectiveness of the proposed set of features. The maximum classification accuracy was 89.29% while the maximum mutual information (MI) obtained-the competition criterion-was 0.70 which outperform the state of the art algorithms proposed for the same dataset.