In most astronomical observations the image intensity consists of contributions from both object and sky background. Depending on the complexity of the field and the type of object, different methods are used to estimate and subtract the background intensity. For linear detectors this can be done directly on the intensity calibrated frame while special consideration must be given when a non--linear response transformation is used (i.e. for photographic emulsions) due to the non--gaussian noise distribution. In the latter case a fit is normally done to the original data and the fitted values then transformed to intensities and subtracted.
An accurate determination of the background is extremely important for
the latter analysis. Therefore, one prefers to use all pixels, which
are not contaminated by sources, and fit a low order polynomial
surface to the background. Non--linear filters are often used to
remove stellar images and other sharp features (see
Section 
) while extended objects are very difficult to
eliminate automatically. If only point like objects are analyzed
background following methods like the recursive filter described by
Martin and Lutz (1979) can be used.
Also 
--clipping techniques are applied in an iterative
scheme where pixels with high residual compared to a low order
polynomial fit to the frame are rejected (Capaccioli and Held 1979).
In this method areas containing extended objects can be excluded
before the iteration starts. In Figure , a field with
extended objects is shown with the mask defining the areas to be
omitted in the computations.
Figure: Background fitting with an iterative 
technique:
(A) original, (B) mask of included areas, and (C) fitted background.