A hybrid approach to machine learning annotation of large galaxy image databases

Abstract: Modern astronomy relies on massive databases collected by robotic telescopes and digital sky surveys, acquiring data in a much faster pace than what manual analysis can support. Among other data, these sky surveys collect information about millions and sometimes billions of extra-galactic objects. Since the very large number of objects makes manual observation impractical, automatic methods that can analyze and annotate extra-galactic objects are required to fully utilize the discovery power of these databases. Machine learning methods for annotation of celestial objects can be separated broadly into methods that use the photometric information collected by digital sky surveys, and methods that analyze the image of the object. Here we describe a hybrid method that combines photometry and image data to annotate galaxies by their morphology, and a method that uses that information to identify objects that are visually similar to a query object (query-by-example). The results are compared to using just photometric information from SDSS, and to using just the morphological descriptors extracted directly from the images. The comparison shows that for automatic classification the image data provide marginal addition to the information provided by the photometry data. For query-by-example, however, the analysis of the image data provides more information that improves the automatic detection substantially. The source code and binaries of the method can be downloaded through the Astrophysics Source Code Library.