Spatially-Coherent Non-Linear Dimensionality Reduction and Segmentation of Hyper-Spectral Images (PREPRINT)
MINNESOTA UNIV MINNEAPOLIS INST FOR MATHEMATICS AND ITS APPLICATIONS
Pagination or Media Count:
Non-linear dimensionality reduction and vector segmentation of hyper-spectral images is investigated in this letter. The proposed framework takes into account the nonlinear nature of high dimensional hyper-spectral images, and projects onto a lower dimensional space via a spatially-coherent locally linear embedding technique. The spatial coherence is introduced by comparing individual pixels based on their local surrounding neighborhood structure. This neighborhood concept is also extended to the segmentation and classification stages using a modified vector angle distance. We present the underlying concepts of the proposed framework and experimental results showing the significant classification improvements.