Star Alignment issue

[boelaars]

Hello,

I am trying to Star Align a bunch of images, but am having a curious problem. Normally this works without any problem, even on crappy data like this at hand here. Same result on PixInsight 1.8.7 and 1.8.8, on Mac.

Find these 2 frames here: https://drive.google.com/open?id=1bJPGL91vwuWSL9z2o9Q0st8cjbW0jVHo
They're identified by numbers 1 and 6.

Frame 6 is roughly speaking shifted about 30 pixels to the right. A very visually noticeable amount. However, when I run star alignment, using Frame 6 as the reference and all settings at default, it outputs the below for Frame 1. As can be seen in the output, while it says it registered the frames, it says the displacement is only a fraction of a pixel. As a result when integrating them all stars show up double.

This is a result I'd never expect. I'd either expect it to align them properly, or to error out and say it can't make a match as it usually does when I feed it data that's too bad.

Why is that not happening here? (And can it be solved?)

Code: [Select]

Loading target file:
/Volumes/Drobo Mini/temp/d/Orion_120sec_1x1_all_frame1_d.xisf
Loading image: w=5202 h=3465 n=3 RGB Float32
3 image properties
58 FITS keyword(s) extracted.
Structure map:
Detecting stars: done
Fitting 547 stars:
236 PSF fits.
Matching stars:
171 putative star pair matches.
Performing RANSAC: done
134 star pair matches in 7 RANSAC iterations.
* Summary of model properties:
Inliers       : 0.784
Overlapping   : 1.000
Regularity    : 0.933
Quality       : 0.928
Root mean square error:
delta_RMS     :  0.053 px
RMS error deviation:
sigma_RMS     :  0.023 px
Peak errors:
delta_x_max   :  0.253 px
delta_y_max   :  0.132 px
* Projective transformation matrix:
     +0.999991     -0.000002     +0.017307
     +0.000004     +0.999997     -0.011023
     -0.000000     -0.000000     +1.000000
translation   :      0.02 px
translation_x :     +0.02 px
translation_y :     -0.01 px
rotation      :     +0.00 deg
scale         : 1.000
scale_x       : 1.000
scale_y       : 1.000
Generating registered image
Homographic Projection / Lanczos-3 interpolation, c=0.30:
Registration successful.
Writing output file: /Volumes/Drobo Mini/temp/r/all/Orion_120sec_1x1_all_frame1_d_r.xisf
** Warning: Overwriting existing file
Writing image: w=5202 h=3465 n=3 RGB Float32
79 FITS keyword(s) embedded.
5 image properties embedded.

[Juan Conejero]

Hi,

These images can be aligned without problems. Just disable PSF fittings:

- In StarAlignment, open the Star Detection section.
- For the Compute PSF fits parameter, select Never.

PSF fits are disabled by default in version 1.8.8 when distortion corrections are disabled, so you can simply reset the tool if you use the latest version of PixInsight.

The problem with these images is out-of-focus and saturated stars. PSF fits cannot be computed reliably under these conditions, so almost no stars are being detected if this option is enabled. Since there are hot pixels in the images (and PSF fits are 'perfect' for them), SA is using them as registration references. Since hot pixels are at the same locations on both images, you get residual translations on both axes.

[boelaars]

Ohhh!
They do indeed register properly now. Mystery solved.
I thought there must be some kind of bug as it should of course just simply fail if it 'clearly' can't align the stars. But data can apparently be so crap that it counts hot pixels as stars, and align those just fine! Hahaha. It's kind of funny :-)
That is a great new learning. Thank you, Juan.