Novel projection pursuit indices for feature extraction and classification: An inter-comparison in a remote sensing application

Projection pursuit (PP) techniques are used to search for statistically interesting low-dimensional projections of complex, high-dimensional data. These projections reveal data structure useful for automatic classification applications. We derive a novel class of PP algorithms, comparing them with known PP algorithms. Texture-based cloud detection in airborne visible/infrared imaging spectrometer (AVIRIS) imagery from the Jet Propulsion Laboratory is provided as a basis for inter-comparison.