Has Star Detection Changed in the new Star Alignment Tool?

[Sean]

I'm returning to a project from a month ago, where I had aligned a number of fairly faint Ha images with the previous version of the Star Alignment tool - no problems. Now, using the latest release of Star Alignment, I can't seem to align them, with I believe the same Star Detection parameters as before. I'm getting "0 Star Pair Matches" messages on half of the Ha images.

I'm continuing to investigate, but I wanted to make sure that nothing has changed in the actual Star Detection algorithms in this new version of Star Alignment.

Thanks,

Sean

[Carlos Milovic]

Not sure if it has beeing already applied, but Juan has rewriten SA, allowing not only to use similarity between triangles, but any number of points. Also he was working on a more flexible algorithm, to deal with local distortions.

[mschuster]

Sean,

The Star Detector has changed. See Juan's post:

http://pixinsight.com/forum/index.php?topic=4010.msg39043#msg39043

This post refers to an earlier released .js implementation, which may not match what is in the new StarAlignment process exactly, but it is likely similar IMO.

IMO, at least on my Ha subs, Juan's new work is a significant improvement, more sensitive, more robust, and faster.

The default values for StarAlignment dialog parameters have changed, and how the implementation interprets these values has changed also IMO.

You might want to experiment with different values, if you haven't already. In particular, try reducing Star Detector > Log(sensitivity).

Mike

[Sean]

Mike and Carlos,

Thanks for the info - I had missed that post from Juan about the new detector.

I'll play around with the StarAlignment parameters today and see if I can get better star detection on my images. They're 10 minute Ha exposures with not a lot of stars, but they did align fine in the earlier version.

Sean

[mschuster]

Sean,

As a quick test I tried StarAlignment on 20 second 3nm Ha subs taken at f/5. StarAlignment worked fine on them, using the default settings.

Normally I plate solve short Ha subs like these to sync my portable setup. MaxIm/Pinpoint detects these weak stars fine also.

Maybe you could post a zip containing several of your subs?

Thanks,
Mike

[Sean]

Hi Mike,

I just tried a comparison with the old version of StarAlignment, which I still have running on my old laptop.

As an example, one of the Ha images has 138 stars detected in the new version, vs 17158 stars in the old version, both using the default parameters. That number (17158) is obviously incorrect and includes a lot of hot pixels and other noise, but the RANSAC iterations start limiting things to fewer brightest stars, and eventually RANSAC gets enough valid star matches and solves the image. In the new version, only 9 out of 14 images are successfully solved - the other 5 never get enough stars. I changed the Log(sensitivity) to its minimum, -3, but saw little or no change in the number of stars detected.

Granted, I'm working with difficult images, but it seems as if there has been some fundamental change in star detection, unless I'm missing something else. I'll keep trying other changes, and may take you up on your offer to look at the images on your system. I can put them in a shared Dropbox folder for you, but let me keep trying first.

Also, did you mean 20 minutes, not seconds for your images? Mine are 10 minutes 3nm at f/8, so fairly dim. I'll use 20 minutes next time!

Thanks for your help.

Sean

[mschuster]

Yes, 20 SECONDS not 20 minutes. These are 20 second 3nm Ha subs shot at the zenith used to sync my mount (aka Hartmut's SyncZ), so star fields are basically random.

I figured that if Pinpoint can solve them then StarAlignment should be able to align them also.

Mike

[Juan Conejero]

Hi Sean,

As Mike has pointed out, the star detection routine is different in the new version of StarAlignment. In fact, I have rewritten almost the entire tool.

The new star detector should work better in general. However, I have seen some problematic images with many "concave" stars (that is, the star's central pixels are not the brightest ones for some reason). The new star detector is stricter to reject these objects because we need more robustness against multiple and crowded stars for the new arbitrary distortion correction routines. Another problem could be that your stars are too elongated and are being rejected.

If you upload the images, I'd like to take a look to diagnose the problem.

[Sean]

Hi Juan,

Thanks - I'd appreciate your taking a look. What's the best way to get images to you? I'll pick out a few that illustrate the problem.

By the way, these are really marginal images in terms of SNR and number of stars, so if the answer is that they can't be reliably aligned, that's OK. In fact, now that I look at them closely, I'm kind of amazed that the old version of StarAlignment works so well! Maybe I'll just keep it around for these weird cases ....

Sean

[Juan Conejero]

Hi Sean,

I'm kind of amazed that the old version of StarAlignment works so well!

Quite often when we make a big step forward we may see performance losses in marginal cases. The new version of StarAlignment is much more robust to false star detections. In the case of these Ha images, the fact that the old SA was detecting thousands of false stars tells you that it was working under atypical conditions, as you have diagnosed.

You can upload a couple images to any public server, but I don't use Dropbox. Don't you have an Endor account? If not, you can write us requesting one.

[Ioannis Ioannou]

Similar here Juan, after the new Star Detection, I can not plate solve any of my old images with the ImageSolver script - something that was very easy with the previous version (and in my case I did not had thousand of stars detected before).
Now the ImageSolver script fails with zero to 5 stars detected, no matter how good or bad the star shapes are. Tried the advanced parameters but nothing helped.

[Andres.Pozo]

Similar here Juan, after the new Star Detection, I can not plate solve any of my old images with the ImageSolver script - something that was very easy with the previous version (and in my case I did not had thousand of stars detected before).
Now the ImageSolver script fails with zero to 5 stars detected, no matter how good or bad the star shapes are. Tried the advanced parameters but nothing helped.

Hi Ioanis,

is your image in RGB? If it is a RGB image, please, could you extract the luminance and try ImageSolver on it? I detected yesterday some problems in ImageSolver in RGB images with the latest version of the script and the latest versions of the modules of PI.

Andrés.

[Ioannis Ioannou]

Tried it, no help, BUT, I think I narrowed it a lot. I took out my raw images backup and tried to many of them to find a pattern.  Apparently raw images are working, RGB or L. The problem appears with already stretched images, especially in non-fits format. RGB or L does not make a difference.

(In which case my report maybe should be considered as lower priority)

[Andres.Pozo]

Tried it, no help, BUT, I think I narrowed it a lot. I took out my raw images backup and tried to many of them to find a pattern.  Apparently raw images are working, RGB or L. The problem appears with already stretched images, especially in non-fits format. RGB or L does not make a difference.

(In which case my report maybe should be considered as lower priority)

Could you attach the log of the process so I can see where it fails? If you could send me one of the images it would be better?

In any case, ImageSolver has always had problems with stretched images because it its based on StarAlignment and DynamicPSF which are designed for working on linear images. I should write the user manual for this script and include a section of tips.

Meanwhile you can read this message:
http://pixinsight.com/forum/index.php?topic=3911.msg29420#msg29420

[Ioannis Ioannou]

Understood and duly noted. 

(although I have to admit that with the previous version the solver always worked for me - but never mind, if it is only on stretched images it is minor).

I'll send you a PM with an example.