A spatial-spectral approach to deriving eigenvectors for remote sensing image transformations

Spectral decorrelation methods are commonly used in remote sensing to derive eigenvectors that best represent the spectrally distinct materials of a given scene. Separating eigenvectors related to signal as opposed to noise is a difficult task, particularly as image data increases in size. In this paper a novel spatial-spectral approach to eigenvector derivation is presented that can speed up processing, be applied to very large or mutiple image data sets, derive eigenvectors that represent the spectral diversity of the data, and also improve the separation of those eigenvectors representing signal as opposed to noise. These advantages are demonstrated using the well known AVIRIS Cuprite imagery.