Two-dimensional sparse preserving projection for face recognition

The basic idea of sparse representation is that any sample can be accurately reconstructed by few related samples, however, one-dimensional sparse projection damages the structures of samples when turning image matrices into vectors in feature extraction and it also results in the problem of singular covariance matrix. This article extends one-dimensional sparse preserving projection to two-dimensional feature extraction area, and develops a new method called two-dimensional sparse preserving projection (2D-SPP), 2D-SPP can effectively extract the features and resolve the small sample size problem. The experimental results on face database verify the validity of this method.