Quote:
Olivier Hainaut: ESO wrote
Combining the data over a huge wavelength range UBVI can be counter productive, even if your only goal is to go deep. For instance, the individual U images tend to be much shallower (because the camera is less sensitive in U, and bc the stars are fainter in U), so if you average them with the others without a different weight, you actually increase the noise more than the signal. You could consider to use the following weight for each image:
w = F/t
with t = exposure time (i.e. normalize by exposure time) and
F(U) = 0.2; F(B) = 0.4; F(V) = 1; F(R) = 1.5; F(I) = 0.8
|
First of all I was mistaken that the ESO image is a total 31 hr exposure, but a much shorter 7.5 hrs.
What is remarkable however is that according to the FITS headers an extraordinary number of subexposures were taken through each filter but only a small number were selected for stacking.
According to the FITS headers:
U filter 132 subexposures taken best 13 selected for combining.
V filter 1192 subexposures taken best 13 taken for combining.
B filter 873 subexposures taken best 13 taken for combining.
I filter 141 exposures taken best 13 taken for combining.
Subexposures varied from 300s -1000s depending on the filter.
I took up ESO's advice to normalize the filter images according to the weight function based on exposure times.
The noise values were calculated on 2 differently processed images.
(1) Filter images not normalized, pixel mapping applied to combined image.
Noise value:- 6.98 X 10^4.
(2) Filter images normalized, pixel mapping applied to combined image.
Noise value:- 6.03 X 10^4.
There was a 14% reduction in noise when using normalized filter images.
I used Pixinsinsight's noise calculation script.
It confirms that noise is increased for stacking individual images that are not normalized.
Regards
Steven