Experiments about the Generalization Ability of Common Vector based Methods for Face Recognition

This work presents some preliminary results about exploring and proposing new extensions of common vector based subspace meth- ods that have been recently proposed to deal with very high dimensional classification problems. Both the common vector and the discriminant vector approaches are considered. The dierent dimensionalities of the subspaces that these methods use as intermediate step are considered in dierent situations and their relation to the generalization ability of each method is analyzed. Comparative experiments using dierent databases for the face recognition problem are performed to support the main con- clusions of the paper.