Artifacts at bright stars

G

gottsch

Well-known member
Hello All,

I am tracing back artifacts at bright stars that I see most notably in the final image after deconvolution (BlurXT). Going backward through the flow I see them already very early appearing in an RGB channel re-align step. Then progressively getting stronger in following steps.

These are the steps:
  • Full Integration flow with WBPP (no drizzle) (OSC camera ASI2600MC 40x5min subs ackquisition with NINA)
  • RGB channel separation, registration on green, RGB combination (to reduce atmospheric dispersion)
  • Background flatten (GraXpert, no image shown)
  • SPCC
  • Color saturation #1 (entire image)
  • Color saturation #2 with a star mask generated from StarXT (only enhance saturation on stars)
  • BlurXT
ngc253_Starartifacts_Flow.png


It's noticeable that the root of the artifact is starting very early and then only gets progressively worse.

Are there ideas / proposals to improve the flow to avoid / reduce the artifacts?

Thanks,
Gert
 
looks like some kind of a diffraction artifact, besides the secondary vane - what kind of telescope are you using?
 
pfile said:
looks like some kind of a diffraction artifact, besides the secondary vane - what kind of telescope are you using?
Click to expand...
This is a 10inch Skywatcher Newton. I am not referring to the diffraction spikes nor the artifacts of the mirror clamps. Maybe it's not coming out really well in the screen copies. I refer to the circular / arc-shaped artifacts very close to the core of the star.

Clear Skies,
Gert
 
ah ok, well that kind of stuff is usually either an interpolation artifact, or is the result of oversharpening a saturated area of the image.

i guess you can try drizzle to see if that helps. or move blurXT earlier in the process, so you're applying it to a linear image.

rob
 
The screenshots simply don't have the resolution for this. Could you upload full images and post a link?
 
pfile said:
i guess you can try drizzle to see if that helps. or move blurXT earlier in the process, so you're applying it to a linear image.

rob
Click to expand...
The BlurXT is in linear image.

I'll upload actual data when I get back from the office.

Cheers
Gert
 
The screenshots do indeed show atmospheric dispertion being significantly corrected by separate RGB alignment, so the general workflow looks good.
 
fredvanner said:
The screenshots simply don't have the resolution for this. Could you upload full images and post a link?
Click to expand...
I have uploaded the ZIP file with the previews in xisf format.

Flow steps referring to file names in the ZIP are according to the screen capture below.
From top to bottom left column.
Then top to bottom right column.

I can start to see some 'jaggedness' in the bright stars, esp. the core's upper right quadrant even in the first image which is straight from integration (crop). The following steps seem only to emphasize the issue, with de-convolution doing the worst damage.

What do you think to resolve them?

Star_artifact_flow_m.jpg


Thanks & Clear Skies,
Gert
 
you might try masking the star cores during deconvolution. if you were using the PI deconvolution process you can also fiddle with the deringing controls, which BXT does not have. i haven't looked at the images yet, but from the screenshots i don't really see the harsh artifacts until BXT was applied.

you could use SXT to create a star image and then use morphologicaltransformation to shrink down the stars a bit so they only strongly mask the cores. don't forget to stretch the star image though if it was extracted from a linear image.
 
The fact that the ring artifact is visible at the very first stage suggest that it is a registration interpolation problem (that is enhanced at each steps of the processing).

As @pfile suggested in post #4 I think if you drizzle x1 your image it may take care of the problem. You could also try to lower the clamping threshold of the StarAlignement process.
 
Last edited:
nico1038 said:
As @pfile suggested in post #4 I think if you drizzle x1 your image it may take care of the problem. You could also try to lower the clamping threshold of the StarAlignement process.
Click to expand...
I need more help to understand this. What is exactly drizzle 1x?? Isn't that a no-op?? I understand if the drizzle factor is >1 we are upscaling the image. But what is 1x really doing here and why do we expect that it will help?? I'm confused.

Cheers,
Gert
 
gottsch said:
I need more help to understand this. What is exactly drizzle 1x?? Isn't that a no-op?? I understand if the drizzle factor is >1 we are upscaling the image. But what is 1x really doing here and why do we expect that it will help?? I'm confused.

Cheers,
Gert
Click to expand...

drizzle does not use interpolation and thus is immune to the ringing artifacts that can occur during interpolation. 1x isn't a no-op in this sense; it's a different algorithm for registering images.

rob
 
pfile said:
drizzle does not use interpolation
Click to expand...
Well...
There are some senses in which drizzle definitely avoids interpolation. CFA debayering interpolation, in which the signal value of an unsampled pixel is estimated by interpolating between adjacent pixels, is completely avoided by drizzle (at any scale).
When drizzling a mono image it is much less clear, and x1 drizzle of a mono image with a square kernel and a drop-shrink of one is effectively identical to the standard image registration interpolation (any differences will be in exactly how distortion mapping is handled). This image (from a paper by Fruchter and Hook with a good summary of the drizzle technique) may make this clearer:
1699998637498.png

If the drop size is the same as the original pixels (drop size = 1: so blue squares are the same size as the red squares), and the scale is x1 (so the fine output grid is the same size as the input grid), this just becomes standard alignment interpolation.
 
You must log in or register to reply here.
Back
Top