Photometry based gradient removal

A

AstrGerdt

Well-known member
Hi,

while I was working on my last set of images I was having many problems with heavy background gradients and a lot of IFN which made it near impossible to distinguish between gradients and real signal.

Thinking about my problem, I wondered if it could be possible to remove gradients in a photometry-based way. Let's take the following example:
My image contains two stars, A and B. A is 1.1 times as bright as B.
Now I look up the stars in a catalog and find out, that A is in reality exactly as bright as B.

So in my Image must be a gradient that causes B to be 10% brighter.

Doing this with the ratio of enough stars in the image and the real data from a catalog would lead to a pretty accurate description of gradients in the image. This would remove the problem of distinguishing between real signals and gradients and could lead to a perfect gradient removal.

CS Gerrit
 
i think both LN and NormalizeScaleGradient kind of do this, but i'm not sure how "complex" the generated gradient is allowed to be.
 
As far as I understand, LN and NSG do not actually remove gradients. They just "adapt" the gradients in an image to match the gradients of a reference image, so it becomes easier to remove them later in processing.

What I was talking about was using photometry to actually remove the gradients and not just simplify them or adapt them to a reference image.

CS Gerrit
 
Unfortunately, gradients don't work that way. Additive sky gradients cannot be modeled by measuring the brightness of a set of stars. Multiplicative gradients could be modeled this way, but these gradients are very subtle under normal conditions and they are virtually never what you want to fix in real images. Star catalogs provide no information that can be used to compare local background levels measured on the image to the actual sky local background at each star location (e.g., from data acquired without atmosphere, such as Gaia catalogs). Long story made short: photometry by itself does not provide information useful for effective gradient modeling.

If you want a *really* accurate and reliable method for gradient correction, no software solution can provide it. You need methods based on observational data. Here is one that we have developed:

PixInsight — Multiscale Gradient Correction

PixInsight - Advanced Image Processing. PixInsight is a modular, open-architecture, portable image processing platform available for FreeBSD, Linux, Mac OS X and Windows.
pixinsight.com pixinsight.com

Having said that, photometry *and* spectrometry together could in theory provide sufficiently accurate information for gradient modeling. We are currently working on new tools that use Gaia DR3 mean spectrum data. We'll probably release these tools (sorry but I cannot disclose more information at this point) by the end of August, along with a new set of XPSD database files with Gaia DR3 data, including BP/RP sampled mean spectra for about 200 million stars (be prepared to download about 70 GB of database files). Of course we'll explore this possibility.
 
  • Like
Reactions: dld
Once I had the same thought and I was ready to experiment with it by using some python tools. Nowadays I am considering another approach, which I will not reveal at this moment.

I am not sure if it is possible to use equal magnitude stars as reference for such task: Light pollution is an additive component which means both the peak and the tail of a star's profile will be equally affected. In theory, photometry can be performed equally well under light pollution. In other words, the excess sky flux due to a star should be the same, irregardless of LP.

I believe what LN does is to seek relative changes in brightness and adapt a target frame to match a reference frame: For each star in the target frame the star's local sky brightness is computed and compared to the local sky brightness of the same star in the reference frame. By subtracting a function which models this excess local sky brightness for each star, the target frame is now comparable to the reference frame. Given that, I don't know what can be achieved if the reference image is a survey image (thus a reference image with non-existent or accurately-subtracted light pollution).

(I hope I have a correct understanding of the core principle behind frame normalization!)
 
Thanks for your reply, Juan. As always, great work and knowledge! I'm already looking forward to the seemingly upcoming update.

@dld I was also thinking about using LN against a "perfect" image of the region. The problem is, how to get this perfect image? Taking it during the night, as in the multiscale gradient reduction approach, is impossible for me, due to restrictions in the field of view. There are just too many buildings around.
I thought about artificially generating the star field, but that would lead to other problems due to the more or less random setting of the star brightness.

CS Gerrit
 
@AstrGerdt honestly, I don't know. The "perfect image" might be some kind of a professional survey image but even if I had access to raw linear data, I don't know if such images are compatible for the task.
 
I am pretty sure survey images would work in terms of being compatible. If they are actually sufficient to remove gradients is another question.

For example, the DSS images often show what I would consider a very severe gradient. These could of course not correct any gradients.

CS Gerrit
 
You must log in or register to reply here.
Back
Top