For projects which perform statistical analysis on groups of objects, it is important to rely on an objective search method to extract them from the data frames. For this reason plus the need to search large areas efficiently it is necessary to use automatic algorithms for this task. Several different methods are applied for this purpose depending on the demands for speed, limiting magnitude, and location of special objects. They fall in four main categories depending on their detection criterion, namely : level, gradient, peak, and template match detection.
The simplest and fastest method is using a given level over a
previously determined background value as the criterion to identify
possible sources. All pixels with intensities over this value are
flagged and later grouped together to form objects (Pratt 1977). The
background estimation can be avoided by using a gradient of
the intensity distribution (Grosbøl 1979) instead. If the background
variation over small areas can be regarded as linear, a Laplacian
filter (see Equation 
) will locate only sharp
features such as edges of stars. Since the derivative has a larger
noise than the original image, this method will be slightly less
sensitive than using the level. It can be applied directly to data
without first having to compute the background and may therefore be
faster to use if only point sources should be detected. The peak
detection method finds pixels which are higher than their surroundings
and is also based on a derivative (Newell and O'Neil 1977; Herzog and
Illingworth 1977). Especially in crowded fields where a background is
difficult to determine and where objects may overlap, it is a better
search criterion than the two previous schemes. Finally, it is
possible to compare each position with a template (e.g. the PSF) and
thereby determine the probability of having an object there. Although
this gives the most sensitive search criterion because it uses all
information, it requires much larger amounts of computer time than the
other methods.