Change Detection with Manifold Embedding for Hyperspectral Images

This paper proposes a manifold based approach for change detection in multitemporal hyperspectral images. Manifold representation, using Laplacian Eigenmaps, is applied for dimensionality reduction on stacked temporal datasets and change detection on the reduced datasets. The resulting latent vectors are utilized to cluster the changed vs. unchanged regions. A semi-supervised scheme is also proposed which circumvents the challenging thresholding issue and enables satisfactory binary change detection outputs. The proposed approach is validated on two real bitemporal hyperspectral datasets.