Back to forums

Alignment difficulties

Pleiades Astrophoto PixInsight Processing Registration Deep Sky NGC Objects
7 replies•294 views
Steve Solon avatar
Steve Solon avatar
Hi imaging folks,

I've never had ANY concerns when aligning images in PixInsight - until now, and I'm not sure why. I'm trying to align 233 three-minute subs centered on NGC-3190 in Leo. There is some slight mis-alignment among several of the 233 subs, but nothing drastic. According to PI, there needs to be at least 6 star matches for alignment to succeed. Out of the 233 subs, 27 fail. I've looked at the 27, and yes, there is some slight differential distortion, but again, nothing to worry about - there are at least 6 stars common to the images that WILL align. The attached screen shot is an example of the Reference Image for alignment, and the settings I'm using, but am still getting fails. I've tried both Projective Transformation and Thin Plate Splines, with the same results.

I'm running Windows 10 with 32gb of RAM. Any suggestions? This the first time there's been an alignment concern. I've actually aligned other target images with fewer stars, so this one is a head-scratcher. I'd hate to lose 81 minutes of data without understanding why.

Many thanks for your thinks.
 - - Steve
Well Written Engaging
kuechlew avatar
kuechlew avatar
I actually went from DSS to PI due to its very reliable alignment routines and never had it failing on me.

Looking at your image I realize that there are very few stars and the obvious ones are either saturated or a little bit distorted. Please have a look at the logs first. StarAlignment will tell you how many stars it finds in the reference image. Depending on this number you may want to play around with the star detection parameters and/or the reference image. The logs will tell you too how many matching star pairs are found. Obviously if PI finds has difficulties detecting stars the chance to detect matching star pairs is vastly reduced. For details on what parameters to tweak you may have a look at PixInsight Reference Documentation | StarAlignment
On a first glance it seems reducing "detection scales" may be the first step to take in case finding stars is an issue.

Once you have the star detection resolved (!), in case the problem still persists, you may try to tweak the star matching parameters:
- Changing the descriptor type may help. It should be easier to find matching triangles than matching hexagons, you may try with quadriliterals or pentagons first.
- Increasing "Descriptors per star" may help to find matching polygons at the cost of processing time. The more polygons the algorithm looks at the higher the chance to find a match.
- Increasing the RAMSAC tolerance value may help too but given your description I doubt that's the problem.

Hope this helps a bit.

Good luck and clear skies
Wolfgang
Helpful Engaging Supportive
kuechlew avatar
kuechlew avatar
One additional idea: Depending on how large the misalignment between the images setting "Compute intersections" to always may be necessary to determine the common area between the images.

Clear skies
Wolfgang
andrea tasselli avatar
andrea tasselli avatar
Use DynamicAlignment.
Bruce Donzanti avatar
Bruce Donzanti avatar
Leave most of your settings as they are, but in addition, try the following setting changes:

Star Detection
Noise Scales: 1
Noise Reduction: 1

Star Matching
RANSAC tolerance: from 6 to 4
RANSAC iterations: 4000
Descriptors per star: 80

When I want to do a quick and dirty test of my data before doing complete pre-processing, I use these settings which work even on faint objects with few stars in the FOV.
Helpful
Steve Solon avatar
Steve Solon avatar
Wolfgang, Andrea, and Bruce - 

Your suggestions are much appreciated, and expand my knowledge of PI. I will certainly try them all, and let you know the results. 

- - Steve
Well Written Respectful
Bob Lockwood avatar
Bob Lockwood avatar
Not sure if this helps or not as I don’t use PI. What I do use for all my calibration is from CCDware, the program is CCDStack, yes it’s another program to buy, not cheep $200USD and the plugin called CCDIS/P for another $90USD is needed. If you get or have there CCD Inspector in addition to CCDStack, it comes with CCDIS, but that’s beside the point. In CCDStack with CCDIS, you open all your already calibrated subs, you know, darks, flats bias exc, that have already been done and you go to Stack-Register, and there all you do is hit align all. Now I don’t know about 233 frames as I usually never have more then 50 or so max, but it will pick a reference frame, look at between 200 and 300 stars, it will match anywhere from 20 to 75 or so stars and align everything and it doesn’t mater if it’s rotated, flipped from a pier flip, or even binned, it will align everything back to a 1x1 and to within 1/10 of a pixel as per CCDStack. Works quite well.
Steve Solon avatar
Steve Solon avatar
Good info, Bob - thank you. Many years ago, I used a wonderful program from Santa Barbara Instruments Group called CCDSoft - it did everything except wash the dog, and I'm sure somebody could have written a plug-in for that!. I have heard of CCDStack, and will give it some thought - the suggestion is much-appreciated!

- - Steve
Well Written Respectful