Questions about the solving process (verification, background subtraction, distortion correction,..)

[Maik Riechert]

Hi,

I'm trying to understand how astrometry.net really works and have some questions on that. My setting is ISS images where the lower half of the image is the earth and in the upper part there are the stars and maybe some spacecraft structures like solar panels. I am already masking out all the non-star areas (that is, painting them black).

Question 0: What is the difference between 'distractors' and 'conflicts'?

Question 1: Suppose that 80% of the image is masked. When astrometry.net finds a quad candidate, it is looking for other matching stars in the image (verification of hypotheses). Is it only considering stars which have been previously extracted? Or is it looking for possible stars anywhere in the image, in my case in the black masked areas? If the latter is true, would the masking disturb the verification phase dramatically and maybe lead to it throwing away solutions which might be fine if only the non-masked areas had been checked?

Question 2: To solve these images, I had to disable background subtraction for star extraction. This makes sense as the subtraction would get confused by all the totally-black areas when subtracting the median of the image and also when computing the noise level. Is that correct? If yes, is there a way to still do the background subtraction for such images to improve the star extraction? I guess astrometry.net would need to be given the mask then to be intelligent about it.

Question 3: I read the astrometry.net paper from 2010 and it doesn't mention the tuning and SIP corrections available in astrometry.net. Is there any new documentation on that or is it still considered very experimental?

Question 4: Which softwares exist that can fine-tune astrometry.net's WCS solutions, supporting optical distortions? I only know about SCamp but haven't tried it yet as it seems to be made for aligning multiple overlapping solutions.

Thanks a lot for answering these, it is very much appreciated!

Cheers

Maik

[Dustin Lang]

Hi,

Question 0: What is the difference between 'distractors' and 'conflicts'?

A 'distractor' is a star in your image that wasn't expected -- that isn't in the reference catalog.

A 'conflict' is when two stars in your image are close to a reference star.

 

Question 1: Suppose that 80% of the image is masked. When astrometry.net finds a quad candidate, it is looking for other matching stars in the image (verification of hypotheses). Is it only considering stars which have been previously extracted? Or is it looking for possible stars anywhere in the image, in my case in the black masked areas? If the latter is true, would the masking disturb the verification phase dramatically and maybe lead to it throwing away solutions which might be fine if only the non-masked areas had been checked?

I always forget which way around this goes......  Given a matching quad, we search for reference-catalog stars that should be in your image.  Then we go through the stars in your image (in brightness order), asking which ones match to the reference stars.

There is some cost for having no stars in parts of the image -- we will find reference stars there, and the probabilities are normalized by the number of reference stars, so if there are ones you can never hit, it hurts a little.

Generally, the "log-odds" -- the relative probability that it's a correct match vs a random match -- is huge, so this usually doesn't matter much.

| Question 2: To solve these images, I had to disable background subtraction for star extraction. This makes sense as the subtraction would get confused by all the totally-black | areas when subtracting the median of the image and also when computing the noise level. Is that correct? If yes, is there a way to still do the background subtraction for such | images to improve the star extraction? I guess astrometry.net would need to be given the mask then to be intelligent about it.

Yes, large areas of solid black will throw off the background estimation (not too much, I wouldn't have thought) and noise estimation (definitely!).

Maybe instead of filling the blank areas with solid black you could fill it with something like the average background value, if there is such a thing, plus random Gaussian noise of the typical noise level.  Or specify the noise level (solve-field --sigma).

> Question 3: I read the astrometry.net paper from 2010 and it doesn't mention the tuning and SIP corrections available in astrometry.net. Is there any new documentation on
> that or is it still considered very experimental?

I don't think there's any real documentation on it.  What's implemented now is not a great algorithm.  It uses a kind of "simulated annealing" approach.

> Question 4: Which softwares exist that can fine-tune astrometry.net's WCS solutions, supporting optical distortions? I only know about SCamp but haven't tried it yet as it
> seems to be made for aligning multiple overlapping solutions.

I only know of Scamp.

Ages ago I started implementing a feature that I think might help for images with significant distortion.  The idea is to learn a SIP distortion solution (hand-chosen from a good solution, probably), apply its inverse before solving (so that the solver can work in undistorted coordinates), and then re-apply the distortion to the solution.  I never finished that, though...

--dstn

[Maik Riechert]

Dustin Lang wrote:

Ages ago I started implementing a feature that I think might help for images with significant distortion.  The idea is to learn a SIP distortion solution (hand-chosen from a good solution, probably), apply its inverse before solving (so that the solver can work in undistorted coordinates), and then re-apply the distortion to the solution.  I never finished that, though...

 

This sounds good! Instead of hand-choosing a good solution, you could even automate this in case there is a lens profile available for the lens given in the EXIF headers. See http://www.digitalcameraworld.com/2013/12/04/lens-correction-tricks-for-photoshop-how-to-correct-distortion-and-other-optical-flaws/. 

Hmm, thinking about it, in my case I don't care about how the original image looks like because I project it onto earth afterwards. This means I could just correct the distortion in the images before-hand myself, and then use astrometry as usual, in the hope that the solutions are better. A good plan!

[Maik Riechert]

> Question 4: Which softwares exist that can fine-tune astrometry.net's WCS solutions, supporting optical distortions? I only know about SCamp but haven't tried it yet as it 

> seems to be made for aligning multiple overlapping solutions.

I only know of Scamp.

 

I just tried to use the --scamp-ref parameter but scamp then complains:

----- SCAMP 2.0b15 started on 2014-01-21 at 17:54:01 with 8 threads

----- 1 inputs:

> Examining Catalog image.scamp

> *Error*: XWIN_IMAGE parameter not found in catalog image.scamp

Is this a problem in astrometry.net or am I doing something wrong?

Cheers

Maik

[Dustin Lang]

SCamp needs a reference catalog (eg, stars from USNO-B or 2MASS), plus a catalog for each image you want to align.

solve-field --scamp-ref ref.scamp --scamp image.scamp

Creates those two files, respectively.

You might want to do:

solve-field --scamp-ref ref.scamp --scamp image.scamp --scamp-config scamp.conf

To get a config file that should tell SCamp to use those two files.

It has been years since I have done this, so my memory is foggy on the details.

--dustin