A computer eyes the heavens.

Like wine tasters with world-class palates, astronomers trained in the art of classifying galaxies by shape can scarcely keep up with the demand for their services.

To distinguish elliptical galaxies from spirals and the myriad shapes in between requires years of tedious practice. A glut of astronomical data has overwhelmed the few experts capable of such "morphological" classification.

To help with this problem, Ofer Lahav of the Institute of Astronomy in Cambridge, England, and a team of astronomers and computer scientists have trained a computerized electronic eye in the subtleties of galaxy classification.

"The challenge is to design a computer algorithm that will reproduce classification to the same degree that a student or colleague of the human expert can," the authors explain in the Feb. 10 Science. For their test system, they chose an artificial neural network. Originally suggested as simplified models of the human brain, such networks "are computer algorithms that provide a convenient general purpose framework for classification."

The automated procedure involves two steps. First the system extracts key features, such as a galaxy's color or number of spiral arms, from a digitized image. Then it fits those features into a classification, based on a "training set" of images organized by a panel of astronomers.

Overall, the neural network proved a bright student, the authors contend. Classifying 831 galaxies, the computer deviated from a panel of six expert observers only 9 percent of the time. That's not so bad, considering that panel members themselves agreed unanimously on only 8 of the 831 galaxies.

"Our results indicate that the [neural network] can replicate an expert's classification of the [galaxies] as well as the colleagues or students of the expert," the scientists conclude.