Statistical Feature Recognition for Multidimensional NSO Imagery

Turmon, Pap, and Mukhtar (2002: Astrophysical Journal 568, 396) present a statistical method for identifying sunspots, faculae, and quiet Sun region classes in co-registered SOHO/MDI magnetograms and intensity images. This paper describes progress toward an extension of this method for finding a more complete region classification using multidimensional images (magnetic flux, line-of-sight velocity, intensity, equivalent width, and central line depth) obtained from 1992-2003 with the NASA/NSO Spectromagnetograph (SPM) and since 2003 with the NSO/SOLIS Vector Spectromagnetograph (VSM). We discuss the selection of the feature set, training images, and the temporal and spatial consistency of the NSO data. We determine class-conditional probability densities using both Gaussian mixture models and direct histogram interpolation, and compare feature labelings driven by both methods.