Page translated by Claude — switch to Italian to read the original article.
Calibrating astronomical images is one of the key steps to achieving a quality result, and one of the most important steps in this phase is the application of the flat field.
During my PixInsight courses all over Italy, one question, unrelated to the use of the software but of fundamental importance for image calibration, is: how many ADU should I make my FLATs?
My answer is always and invariably: “if your sensor is linear, and it is very unlikely that it isn’t, it doesn’t matter: the effectiveness of the flat field does not depend on the illumination level”.
The inevitable follow-up question is: “But then why don’t my flats work?”
The answer that has by now become proverbial among my students is: “I don’t know, but consider that over 90% of flat problems don’t come from the flats.”
My confidence stems from absolute trust in the basic theory of image calibration and in the mathematics that describes it.
If a flat is “done well” it MUST work! and among the parameters that define “done well” the illumination level does not appear (at least as far as effectiveness is concerned; the influence on noise is a different matter, but that is another story).
With this conviction I set about calibrating my latest astronomical images and, to my great dismay, I realized that the flat field was not working and was making the images unusable.
It had never happened before and, to be honest, I had moments of discouragement.
Then I thought about my recurring phrase and began to think about where the problem might be.
For the flat to work, the sensor must be linear: fortunately I had verified the camera’s linearity in order to write my last article.
Then there must be no additive components: this is the primary cause of flats that don’t work.
The flat, in fact, must contain exclusively a “map” indicating what percentage of light is absorbed by the optical train (for example because of vignetting or dust), and therefore exclusively multiplicative components; by “additive components” we mean structures and contributions to the image that add light to the pixels, independently of the multiplicative structure.
For example, a poorly blackened, too shiny adapter, or a badly positioned diaphragm or, more frequently, an infiltration of external light is enough to make the flat stop working properly.
Knowing this, however, I had tried to take good care of the entire optical train.
Another fundamental requirement is that the entire optical train does not change between the acquisition of the flats and that of the actual images: same focus point, same rotation, etc.
And here a light bulb goes off: with my old SBIG ST-2000 XM I had never had problems calibrating images.
Now, however, the new Moravian 16200 weighs three times as much, has a much larger field, and is held in position by a 2" “nose” kept firm by the classic brass ring system with a single small screw.
Could there be a flexure in the optical train?
The flats were acquired with the telescope vertical and the flatbox resting on the dew shield, while the images were taken 30° above the horizon with the telescope almost horizontal.
At first glance the camera would seem perfectly rigid, but even minimal flexures are enough, especially in heavily vignetted systems like mine.
And so the idea: it is well known that, by calibrating a flat field with another flat field, you should obtain a perfectly “flat” image containing only noise.
This, for example, is an excellent way to verify independence from ADU (and indirectly the linearity of your sensor).
So, in order to verify flexure in the optical train, I acquired a flat with the telescope vertical (the one I normally use for capturing these important calibration files), then a second flat with the telescope horizontal, keeping the other conditions unchanged (flatbox brightness, focus point, and exposure time).
I then used the first flat to calibrate the second (using bias frames as flat darks): if the telescope is rigid then I should obtain a perfectly homogeneous image without structures; if instead there were significant flexures (even if invisible to the naked eye), the final image would contain inhomogeneities.
And here is the result of the experiment; evidently my optical system is not as rigid as it seems.
