All of the existed algorithms in face recognition

which can obtain uncorrelated discriminant vectors

are based on image vector model

so they encounter so called “small sample size” problem. These algorithms

which are solved using recursive methods

require much computation time. So a new algorithm is proposed in this paper

which is called Two-dimensional Uncorrelated Discriminant Vectors based on an image matrix model. The new algorithm solves small sample size problem through whitening transform of within-class scatter matrix

which makes the model of extended Two-dimensional Linear Discriminant Analysis have similar form of Two-dimensional Principal Component Analysis model. Thus two algorithms were combined effectively

uncorrelated discriminant vectors can be obtained non-recursively. The new method computes fast while maintaining numerical stability. The numerical experiments on facial databases of ORL and Yale show that the proposed method has not only reduced the computation complexity but also achieved higher recognition accuracy

providing new thought on how to obtain uncorrelated discriminant vectors.