Optimal classification of images into stars or galaxies - A Bayesian approach
William L. Sebok · The Astronomical Journal · 1979
Bayesian decision theory is applied to the star/galaxy classification problem and a classifier is derived. This classifier can be written in the form of a correlation with a single stored function. The classifier is then applied to images on 48-in Schmidt plates. There are no free parameters which need to be adjusted for each plate. Calibration of the classifier to a plate simply involves obtaining images of (brighter) objects known to be stars. Sources of error are discussed and a second classifier, which is insensitive to variations in the sky background, is derived and applied to the plate data. Finally a prediction of the magnitude limit is derived for both classifiers and compared to the observed magnitude limit. This observed magnitude limit is about one to one and a half magnitudes above the plate detection limit.