Supervised Laplacian Eigenmaps for Hyperspectral Data

With Laplacian eigenmaps the low-dimensional manifold of high-dimensional data points can be uncovered. This nonlinear dimensionality reduction technique is popular due to its well-understood theoretical foundation. This paper outlines a straightforward way to incorporate class label information into the standard (unsupervised) Laplacian eigenmaps formulation. With the example of hyperspectral data samples this supervised reformulation is shown to reinforce within-class clustering and increase between-class distances.