An extension of parametric decision boundary feature extraction (DBFE) for classification of hyperdimensional data

Decision boundary feature extraction (DBFE) estimates the decision boundary between individual classes and uses it for feature extraction. Because the DBFE relies on the estimate of the decision boundary, it fails when not enough data are available. To overcome this problem, it is suggested to increase the size of the training set by including random data based on the estimated mean vectors and covariance matrices of the classes in the original training set. In experiments, this approach shows potential when very limited training data are available for some classes