|
Detection of the Variables
|
In the subtracted images, only the variable stars
should let significant correlated residual. Once the subtracted images
are normalized with the Poisson fluctuation, the variable stars usually
show up very well. Although if you are interested with faint variables,
it is important to stack all the normalized subtracted images in order
to get better detection. It is basically what detect.csh
is doing. However, things are once again complicated by the presence of
possible defects in the image. Cosmics for instance will induce false detections,
it is even likely that there will be more cosmics than genuine variables.
To avoid this problem we need to perform some kind of rejection. Imagine
we look at the same pixel on each subtracted image, we can build a time
series of the different pixel values. If we take the median of this time
series we will have something very robust with respect to the cosmics or
other defects. However, the median can be efficient for periodic variables
but deadly to other ones. To reject the bad points we order the time series
of absolute deviations and find its maximum. This maximum (first larger
deviation) is then compared to the value of the Nth deviations.
If the Nth deviation is less than half of the maximum, it is very likely
that the absolute deviation is driven by a few points: a few defects or
cosmics. Thus in this case the mean of absolute deviation is clipped from
the N largest deviations. Otherwise the mean of the absolute deviations
is taken. You can set the parameter Nth looking for the keyword n_reject
in register2/process_config.
The detection process involve also some smoothing of the images, you can
specify the size of the smoothing mesh by looking for the keyword mesh_smooth.
You can then run detect.csh. It will produce 2
images: var.fits
(mean of absolute normalized deviations) and abs.fits
(mean absolute deviation). Look at var.fits, you will see that the 5 variables
are well visible. If you constructed the reference with ref.csh, you should
see a pattern of residual along the edges of the images. This is normal,
it is due to the fact that convolution have been used 2 times. Once to
build the reference and another time to make the subtraction. In this case
one cannot detect variable at a distance less than about the kernel size
from the edges.
Type: ./detect.csh
By looking at var.fits you should select a threshold
in order to select the variables. Any variable with a maxima above this
threshold in var.fits will be selected. You can specify the threshold with
the keyword sig_thresh.
You can then run find.csh, the output will be
phot.data, which will contain the position of the variables.
Type: ./find.csh
Next:
Getting the light curves