Pixinsight: Registration problems

[Garbz (Chris)]

Hi all, I've been battling with this for pretty much most of the day now. I'm doing a stack in Pixinsight and it seems to throw away a signficant number of my frames during the registration process. The error I get is:

** Error: RANSAC: Unable to find a valid set of star pair matches.

For some reason some of my subs just won't register. The results are dependent on which frame I use as my reference frame. Playing with the settings in the Batch Pre-Processor doesn't resolve the error or number of images causing the error. It does change the number of stars detected but this hasn't been a problem so far with nearly every image coming up with a note saying it's selected the "500 brightest stars".

I've posted an example of two images before. Before anyone points out the obvious that one is upside down, I know . But that doesn't seem to be the problem as subsequent frames after the meridian flip register fine.

Normally I wouldn't fuss about this but it's throwing away 1/4 of my data in this case! I can't see any problem with SNR, in fact it seems to register some 60 second exposures just fine too. I don't see any real problem with star elongation either. Both of these were suggested as possible solutions elsewhere.

Does anyone have any tips or suggestions of what to try next before I try manually stacking? Because that process just sucks.

[RickS (Rick)]

Have you just been using BPP to do the registration, Chris? I've noticed that lately it has been failing on a bunch of subs but if I run BPP only to calibrate and then run StarAlignment manually it works fine. I haven't bothered to figure out why because it's only a minor inconvenience but I'll get around to it some time. If that's your problem then I'll feel slightly more obliged to track it down...

Cheers,
Rick.

[Garbz (Chris)]

Yeah I was using BPP. I guess manual will be the way to go then. I bet it's some default value that BPP uses for star alignment.

yay more to learn. There's only a million options to set in each step

[Garbz (Chris)]

Hey Rick I'm getting no where with StarAlignment. It finds some 1000 stars, selects a bunch of the brightest ones and then starts the RANSAC thingamabob. It finds about 10 putative star pair matches, and then fails and tries again and goes through that about 16 times and then throws an error.

I haven't figured out any setting that will actually let it find a matched pair of stars. I'm beginning to wonder if there is something bizarre in my subs.

If you at all have a moment would you be able to take a look and see if you can find some-way of making these two images align?

http://www.garbz.com/hamfail.fit
http://www.garbz.com/hamref.fit

By the way do you debeyer before registering?

[RickS (Rick)]

Originally Posted by Garbz

Hey Rick I'm getting no where with StarAlignment. It finds some 1000 stars, selects a bunch of the brightest ones and then starts the RANSAC thingamabob. It finds about 10 putative star pair matches, and then fails and tries again and goes through that about 16 times and then throws an error.

I haven't figured out any setting that will actually let it find a matched pair of stars. I'm beginning to wonder if there is something bizarre in my subs.

If you at all have a moment would you be able to take a look and see if you can find some-way of making these two images align?

http://www.garbz.com/hamfail.fit
http://www.garbz.com/hamref.fit

By the way do you debeyer before registering?

Hmmm... works for me? I took the two files, Debayered them both (default parameters) and then ran a star alignment with the ref file as the reference image and the fail file as a target (default parameters again) and it ran fine:

StarAlignment: Global context

Reference view: hamref_RGB_VNG
Structure map: done
Detecting stars: done
533 stars found.

Registering target image 1 of 1
Target view: hamfail_RGB_VNG
Structure map: done
Detecting stars: done
435 stars found.
Matching stars: done
336 putative star pair matches.
Performing RANSAC: done
315 star pair matches in 5 RANSAC iterations.
Summary of model properties:
Inliers : 0.938
Overlapping : 0.924
Regularity : 0.947
Quality : 0.919
Root mean square error:
delta_RMS : 0.266 px
RMS error deviation:
sigma_RMS : 0.114 px
Peak errors:
delta_x_max : 1.318 px
delta_y_max : 0.663 px
Transformation matrix:
+1.0004 -0.0021 -43.0781
+0.0018 +1.0000 +115.1004
+0.0000 -0.0000 +1.0000
scale : 1.000
scale-X : 1.000
scale-Y : 1.000
rotation : +0.10 deg
dx : -43.08 px
dy : +115.10 px
Generating registered image: hamfail_RGB_VNG_registered
Homographic Projection / Lanczos-3 interpolation, c=0.30: done
Registration successful.

===== StarAlignment: 1 succeeded, 0 failed, 0 skipped =====
12.35 s

I can't think of any obvious reason why this wouldn't work for you. I'm running 1.08.02.1098. Maybe you could try exactly what I did and see if you get the same results? I set up the reference and target images as views rather than files. Nothing else magic about what I did.

You must debayer before registration. Registration will move the pixels around and loses the correspondence with the Bayer matrix.

Cheers,
Rick.

[Garbz (Chris)]

Originally Posted by RickS

You must debayer before registration. Registration will move the pixels around and loses the correspondence with the Bayer matrix.

This sounds like something so incredibly obvious that I should crawl into a hole and die of shame.

That worked now. I feel a bit . I'll give it a go on all hundred photos when I get home. Currently I've only got those two images here.

[RickS (Rick)]

Originally Posted by Garbz

This sounds like something so incredibly obvious that I should crawl into a hole and die of shame.

That worked now. I feel a bit . I'll give it a go on all hundred photos when I get home. Currently I've only got those two images here.

A lot of things are obvious only once they have been pointed out

[Garbz (Chris)]

Hey Rick, thanks for your help. I figured it all out. After doing it with Staralign and working on the debeyered images (I thought I was taking the output from the calibration step in BPP, turns out the output was actually in a subfolder called debeyered...) it still failed on 10 images.

When the calibration step was done, the overscan portions of the image was different in each picture. On the long exposure image the overscan area is brighter. Example attached.

I didn't know the significance of it but I batch cropped all the overscan areas out and now I managed to get 100% of the images to stack.

Thanks a lot for your help.

[RickS (Rick)]

Chris,

Glad to hear you got it sorted out. The additional rows & columns usually contain more than overscan. Calibration with overscan regions defined should also remove the extra cruft and just leave the bare image.

Cheers,
Rick.

[Garbz (Chris)]

Man I love online forums. Came here to find out why registration wasn't working, left here finding I've been stacking images wrong all these years. Turns out if you have an overscan in the frame you need to define it so that Pixinsight will remove the overscan component from the bias frame before creating a bias master. This may be why I had those problems with pixinsight inducing noise when stacking last year which was solved by creating some 100 bias frames...

Thanks a lot Rick

[RickS (Rick)]

My pleasure, Chris. We're all still learning