Abstract
A novel approach for decoding the finger flexion and extension from the human electrocorticogram is proposed. First, for different finger movements, we use projected MUltiple SIgnal Classification (projected MUSIC) as a source localization technique to estimate the active areas in the primary motor cortex. Next, in order to distinguish between the flexion and extension, the results of the single-trial-based source localizations are fed as the input features to a classifier for decoding. The performance of different techniques such as Support Vector Machine (SVM), Perceptron, and the k-Nearest-Neighbor (kNN) are investigated and the resulting classification accuracies are 71.59, 79.1, and 86.33 respectively.