conduct non-linear dimensionality reduction, with the assumption
that the high dimensional data lies on a low dimensional manifold
embedded within the ambient space. Locality preserving
projections (LPP) [7] method is a direct linear approximation of
Laplacian eigenmap and shares many of the data representation
properties of nonlinear techniques such as Isomap, LLE, and LE.
LPP finds linear projective subspaces that optimally preserve
the neighborhood proximity structure of the data. Some recent
advances in manifold-based face recognition can be found in
[8–11].
The aforementioned linear discriminant subspace learning
methods such as LDA, UDP and LPP all suffer from the smallsample-size
problem [12], whenever the number of samples is
smaller than the sample dimensionality. In this case, the sample
scatter matrices can become singular in these methods, resulting
in computational difficulty, due to the inversion of a singular
matrix. To tackle this, a separate PCA step is adopted [2,3,7] to
project images from the original image space into a face-subspace,
where dimensionality is reduced to make certain scatter matrices