1 Introduction

Stellar spectral classification is not only a tool for labeling individual stars but is also useful in studies like stellar population synthesis. Extracting the physical quantities from the digitized spectral plates involves three main stages: detection of the spectra, extraction of their images and classification of the spectra. The detection problem and their resolution for digitized objective prism Schmidt plates was presented by Bratsolis et al. ([1998]). The purpose of this paper is to present a fully automated method for the extraction and classification of spectra.

High-quality film copies of IIIa-J (broad blue-green band) plates taken with the 1.2 m UK Schmidt Telescope in Australia have been used. The spectral plates are with dispersion of 2440 Å/mm at $\mbox{H}_{\gamma}$ and spectral range from 3200 to 5400 Å. The photographic material has been digitized at the Royal Observatory of Edinburgh using the Super-COSMOS measuring machine.

A classification problem can be formalized as a pair ( $\cal O,\cal C$) where $\cal O$ denotes a set of objects and $\cal C$ a collection of disjoint subsets ${\cal C}_{1},...,{\cal C}_{l}$ that partitions $\cal O$. The problem is to determine the subset ${\cal C}_j~\subset~\cal C$ to which a given object $o \in \cal O$ belongs. In practice, the set of objects is usually very large, and providing an explicit description of each subset is impractical. The subsets are therefore often implicitly described by specifying a number of typical examples for each subset. Modern computational techniques have been developed to classify large databases of spectra, uniformly and in a considerably short time.

Automated classifiers of stellar spectra have been used in the past. Cross-correlation and minimum distance methods have been originally used by Kurtz ([1982], [1984]). A good review of linear multivariate statistical methods can be found in Murtagh & Heck ([1984]). Stellar classification with artificial neural networks (ANN) as a non-linear technique has been used by many other researchers during the last decade (von Hippel et al. [1994]; Gulati et al. [1994]; Vieira & Ponz [1995]; Singh et al. [1998]; Bailer-Jones et al. [1998]). These methods were utilised for different databases and different spectral dispersion images.

The final stage of our work contains the stellar population synthesis of Magellanic cloud regions with a fully automated method. The detection procedure (Bratsolis et al. [1998]) gives the stellar coordinates on the prism plate. This means that after the classification, we will have a complete mapping of different aged stellar groups of the studied region and this will be the subject of a future study. The low-dispersion objective prism plate was choosen to limit the overlaps between adjacent spectra. We discribe here the extraction method and we compare two simple and effective linear techniques of classification using a sample of 426 low-dispersion extracted stellar spectra from our digitized objective prism plate.