5 Classification and selection

The separation of resolved (= galaxies) and unresolved (= stars, globular clusters, and unresolved background galaxies) was done with the "star/galaxy classifier'' developed by Bertin & Arnouts (1996). It is a neural network trained program that classifies each object with a "stellarity index'' between 0 (galaxy) and 1 (point source). In all fields the separation works fine down to a V magnitude of 21 mag. Beyond this magnitude the classifier values start to scatter. Figure 2 shows as an example the classifier versus V magnitude plots for the NE field (F2) of NGC 1399 and the background field B2. Eye control in all fields showed that all objects with classifier values below 0.35 are well resolved objects, whereas objects above this value can not be clearly classified.

  

Figure 2: The classifier value (0 = galaxy, 1 = point source) is plotted versus the V magnitude for two different CCD fields (see Fig. 1). For V<21 the seperation of point sources and galaxies is obvious. For fainter magnitudes all objects with classifier values below 0.35 ihave been identified as well resolved. Note that most point sources in field F1 are globular clusters that belong to the central galaxy NGC 1399

Therefore all objects were selected that have classifier values below 0.35. Down to a $V_{\rm tot} = 22.0$ and 23.0 mag we found 873 and 1775 galaxies respectively. The properties of all galaxies brighter than 22.0 mag are compiled in a catalog, see Appendix A. This cutoff in the final sample was chosen for several reasons: the finding completeness starts to drop, the scatter in the classifier values increases significantly. Also, at this magnitude we would not find any dEs that follow the surface brightness - magnitude relation of Local Group or Virgo dwarf spheroidals (e.g. Kormendy 1985; Binggeli 1994) due to the limit in surface brightness. See Sect. 7 for a discussion of the completeness of dwarf galaxies in the Fornax distance.

In the background fields B3 and B4 we found with the same selection criteria 668 and 1022 down to V = 22.0 mag respectively. However, one has to be careful when comparing these results with those of the other CCD fields. The pixel size is about 3 times larger than in the first run leading to a 9 times higher area covered by each pixel. We simulated this resolution for two fields of our first run by binning $3 \times 3$ pixel and run SExtractor again. On the one hand, some galaxies have been classified as point sources due to their small angular sizes below the "new'' resolution. On the other hand, some new "galaxies'' have been gained due to the overlap of objects very close in the high resolution image. Down to our magnitude limit of V = 22.0 mag loss and gain of galaxies are nearly balanced and in the order of 8% of the total galaxy counts.