Before the photometric analysis, the raw data are reduced using the MIDAS
environment in the following steps:
bias subtraction, skimming subtraction (for RCA CCD #8), flat-fielding
and cosmic-ray removal.
A "master'' bias frame is obtained by averaging over 
50 frames from which cosmic ray
events have been removed.
As the bias level varies with time, the bias subtraction for each scientific frame
is performed in two steps:
(1) Subtraction of the master bias frame.
(2) Subtraction of the mean difference between the overscan of the scientific
frame and that of the master bias frame.
With RCA CCD #8, the raw CCD frames have a skimming pattern, characterized by
systematic column intensity offsets across the entire CCD.
This pattern varies in intensity at low illumination levels and becomes
constant above 
1000 adu, well below the sky level of all our scientific
frames but above that for the photometric calibration frames.
Because the skimming offsets are additive biases, we can calculate them by using
flat-field frames with different illumination levels. Scaling and subtraction of two
flat-field frames with different exposure times yield a preliminary skimming frame.
As the skimming feature is stable along each column, the signal to noise is improved
by replacing each pixel of a column by the mean value along the column.
The mean level of the entire skimming frame is then adjusted to be zero.
A set of skimming frames derived from pairs of flat-field frames with increasing
illumination levels are calculated. Then for each science frame, the appropriate
skimming frame can be subtracted.
The skimming subtraction is important for the calibration frames where the
sky level is very low. To preserve the quality of our photometry, the calibration
frames for which we cannot adequately subtract the skimming pattern are rejected.
The flat-field frame provides a map of the sensitivity variations over the CCD chip.
This map depends on the spectrum of the incoming light. As the flat-field
pattern is stable from night to night, we obtain for each observing run
a "super flat-field'' in each
filter band by doing a median filtering of all the scientific frames obtained
during the run.
The median filtering removes the objects from the images and yields the
large-scale sensitivity variations. This is the best flat-field frame which can be
obtained, because it is derived from the sky on the science frames themselves.
The number of frames must be sufficient to create a high signal-to-noise final
flat-field frame. In practice, because our fields are sparsely populated,
the super flat-field results from the median filtering of 5 to 20 science frames
(depending on the observing run).
This flat-field is normalized and is divided into each data frame.
The large-scale residual variations in the background of each flat-fielded
frame are 
.
After flat-fielding, we align multiple exposures of identical fields using several unresolved objects, sum the individual exposures and calculate the mean airmass for the final frame.
To remove cosmic rays from the summed frame, we apply a filtering algorithm
kindly provided by P. Leisy. Each pixel is compared with the mean of the
neighbouring pixels. If the pixel value differs by more than 
from the mean, it is replaced by the mean value. The value of 
is large
enough to prevent the subtraction of real objects.