Richard L. White
Space Telescope Science Institute
Automated classification methods are needed for processing the huge quantities of data generated by modern astronomical instruments. The star-galaxy classification problem and some techniques that have been applied to it are briefly reviewed. Methods for constructing training sets and selecting parameters are described.
A new method of scaling parameter values using ranks has been developed. This approach is found to be of great utility for distinguishing stars and galaxies on digitized photographic plates. It should be widely applicable to other classification problems, especially when the data being classified are not completely homogeneous.
Published in Statistical Challenges in Modern Astronomy II, editors G. J. Babu & E. D. Feigelson (New York: Springer), 1997, pp. 135-148. Gzip-compressed postscript version of paper is also available.