problems with StarAlignment under 1.8.7

[airscottdenning]

Trying to align 20 H-alpha subs collected last night, get errors on almost every single frame.

The subs look fine, and are only misaligned a few pixels due to dithering.

This has never happened to me before, and I know StarAlignment has been substantially updated in 1.8.7.

Perhaps I'm missing some new options?

Anybody else having similar problems?

[Juan Conejero]

We cannot help you without a sample of the images. If you upload a reduced set where you are having problems, I'll be glad to take a look.

[airscottdenning]

Wonderful -- will send a Dropbox link later today.

[airscottdenning]

We cannot help you without a sample of the images. If you upload a reduced set where you are having problems, I'll be glad to take a look.

Here's a Dropbox link to a zipped folder:
https://www.dropbox.com/s/9i8abdkeu296od1/PI-1.9.7-SA.problem.zip?dl=0

Contents (6 files):

• 4 images (calibrated H-alpha subs, as .xisf)
• a screen capture of the Star Alignment dialog, and
• the text of the console (errors)

Only the reference image successfully aligned.

Thanks very much for any help!
Scott

[Juan Conejero]

Hi Scott,

Thank you for uploading the data. The new version of StarAlignment performs PSF fits by default for all detected stars. Because of this feature, the new version is much more demanding of good star shapes. Accurate PSF fitting is necessary for the new distortion correction algorithm, which aims at achieving centipixel registration accuracy, even under strong nonlinear distortions.

The problem with these images is that they are out of focus. As a result, PSF fits cannot be as reliable as the tool requires, and hence most stars are being rejected. To solve this problem you have to disable PSF fits, which returns the tool to a behavior similar to previous versions:

- Open the Star Detection section.
- Set Compute PSF fits to Never.

In this way you can register these images without problems. However, if you plan on using the new distortion correction functionality of SA, doing this is strongly discouraged, since PSF fits are necessary to compute reliable local distortion models.

[Andres.Pozo]

Thank you for uploading the data. The new version of StarAlignment performs PSF fits by default for all detected stars. Because of this feature, the new version is much more demanding of good star shapes. Accurate PSF fitting is necessary for the new distortion correction algorithm, which aims at achieving centipixel registration accuracy, even under strong nonlinear distortions.

I don't know if this is a good idea, when the distorsion is so big that it needs to be fixed the stars are usually quite deformed.

I had detected that ImageSolver works worse in the latest version for wide angle images, and this could be the cause. I will have to test disabling the PSF fits.

Andrés.

[Juan Conejero]

In general, except in extreme cases, deformed stars are not problematic for PSF fitting. However, defocused stars can be very problematic due to the uncertainty that a flat profile introduces in the centroid position. This is what has happened in the case shown by the original poster.

For extreme cases however, such as wide fields acquired with typical consumer-grade lenses, disabling PSF fits can be a good idea for plate solving, since in this way you can decrease the number of rejected stars around the corners. For the rest of cases, i.e. reasonable or not excessively bad star profiles, PSF fits are always better than weighted barycenters because they largely reduce uncertainty in star centroid coordinates, which is crucial for the new local distortion algorithm. A new option to enable/disable PSF fits (but enabled by default) would be good for ImageSolver.

Besides this, version 1.8.7 has some new features that can improve plate solving in many and important ways. One of them is a new surface interpolation based on recursive thin plate splines, able to use an unlimited number of interpolation points with time complexity proportional to O(logN):

https://pixinsight.com/developer/pcl/doc/html/classpcl_1_1RecursivePointSurfaceSpline.html

Although this is still experimental, I am using it for local distortions in SA and works very well, even in the current initial implementation. We have a new PJSR object that implements this: PointSurfaceSpline. Let me some time and we'll talk about this in depth.

[airscottdenning]

The problem with these images is that they are out of focus. As a result, PSF fits cannot be as reliable as the tool requires, and hence most stars are being rejected. To solve this problem you have to disable PSF fits, which returns the tool to a behavior similar to previous versions:

- Open the Star Detection section.
- Set Compute PSF fits to Never.

I understand. Thanks very much for your help.

Scott