How to de-ring in v1.9 alpha vs v1.8

[cytan299]

Hi Ivo,
With the new settings in v1.9 alpha deconvolve module, I'm having some problems getting rid of the dark ring around bright stars. Here's the settings:

--- Spatially Variant PSF Deconvolution
Parameter [PSF Resampling] set to [Intra-Iteration + Centroid Tracking]
Parameter [Synthetic PSF Model] set to [Circle of Confusion (Optics Only)]
Parameter [Synthetic PSF Radius] set to [1.0 px]
Parameter [Synthetic Iterations] set to [Off]
Parameter [Spatial Error] set to [1.00]
Parameter [Deringing Detect] set to [21 %]
Parameter [Dyn. Range Extension] set to [1.00]
Parameter [Sampled Iterations] set to [10x]

Here's the result:

v1.9_ring.jpg (5.66 KiB) Viewed 29386 times

Here's what I did in v1.8 which reduced the dark ring around the bright star to more acceptable levels. My settings:

--- Spatially Variant PSF Deconvolution
Parameter [PSF Resampling] set to [None]
Parameter [Synthetic PSF Model] set to [Circle of Confusion (Optics Only)]
Parameter [Sampled PSF Area] set to [15x15]
Parameter [Synthetic PSF Radius] set to [1.5 pixels]
Parameter [Synthetic Iterations] set to [Off]
Parameter [Spatial Error] set to [1.00]
Parameter [Deringing Fuzz] set to [34.7 pixels]
Parameter [Deringing Detect] set to [89 %]
Parameter [Dyn. Range Extension] set to [1.00]
Parameter [Linearity Cutoff] set to [85 %]
Parameter [Sampled Iterations] set to [15x]
Parameter [Deringing Amount] set to [0.92]

and the results:

v1.8_ring.jpg (4.87 KiB) Viewed 29386 times

Is there a recommended setting for v1.8 and v1.9 to reduce the dark ring?

Thanks

cytan

[admin]

Were the same samples used between both versions?

It appears the 1.8 deringing is set rather aggressively (possibly indicating an issue with the samples or settings), however the 1.9 version's point lights do not appear to be as well coalesced as the 1.8 version (possibly due to only 10x iterations vs 15x iterations)...

Would you be able to share the dataset with me?

[cytan299]

Were the same samples used between both versions?

It appears the 1.8 deringing is set rather aggressively (possibly indicating an issue with the samples or settings), however the 1.9 version's point lights do not appear to be as well coalesced as the 1.8 version (possibly due to only 10x iterations vs 15x iterations)...

Would you be able to share the dataset with me?

No, deconvolve samples were different between 1.8 and 1.9 because the "scraggy" boundaries found by 1.9 did not seem to match those found by 1.8.

I'll upload everything this evening: log files, the fit files, masks etc.

cytan

[cytan299]

Hi Ivo,
Please download all the files here (they are zipped):

https://1drv.ms/u/s!ApsRFMpkd9cwjzrZugp ... g?e=aJsq8L

Contents are:
masks.zip: All the masks that I used.
lum_rgb.zip: The LUM and RGB files that were processed.
M63_v1.8.zip: Processed with St v1.8 result: startools_M63.tiff and StarTools.log
M63_v1.9.zip: Processed with St v1.9 result: startools_M63.tiff and StarTools.log


I hope the above will help you diagnose my problem. Here's ST v1.8 processed M63:

startools_M63_small.jpg (481.98 KiB) Viewed 29327 times

The image was taken in Bortle 8 skies over 3 years (yes 3 years) because I can only image on weekends and the sky here is nearly always cloudy.

• LUM data was taken with STF8300C+LPS-D2 filter in 17Apr2020, 19Apr2020 and 29Apr2021 to give 5 hours 35 min

• RGB data was taken with ASI2600MC, no filter on 20 May 2023 to give 1 hour 12 min of data.

It's not the best M63, but I think it's as good as I can do with my processing skills. Hopefully, Ivo can improve on it.

cytan

[admin]

Thank you for uploading that.

This dataset is the absolute worst case scenario for SV Decon.

The noise correlation is extreme; noise clumps/grain are almost as big as the stars;

2023-05-24_14-34.jpg (153.6 KiB) Viewed 29306 times

Did you dither and use bias and/or dark frames? Noise grain should ideally be single pixel in size...

Nevertheless, SV Decon is still reliably identifying possible star samples (yay - I'm so glad I got that sorted now!).

The next problem with thsi dataset, is that the PSF will be... weird. This dataset is the result of multiple nights.
This has caused data to be stacked that varies wildly in PSF (due to seeing and transparency varying). Indeed, this can be seen around the brighter stars where cores from various nights will either over expose (good seeing) or be smeared out (bad seeing). To minimize this problem, I imported the datasets as L, RGB (e.g. I did not use RGB as synthetic luminance), also because combining data from different cameras for the purpose of luminance is never a good idea.

Despite all these challenges, SV Decon does its best. I did not experience any ringing, though I had to set Deringing Detect to 11%.

Selection_788.jpg (61.75 KiB) Viewed 29306 times

Selection_789.jpg (76.9 KiB) Viewed 29306 times

Unrelated, but do check your color balance (for example using MaxRGB mode), as it appears the image you published is rather green. That appears to be because the star highlights contain aberrant color information, likely cause by the mechanism described above (stacking data with different PSFs). This in turn causes the Color module to come up with an incorrect default color balance. Flipping to MaxRGB will readily show the galaxy being green dominant, which is not correct. Reduce the green bias until it no longer is green dominant, and flip back to "Normal" mode. You should now see more familiar galaxy colors.

You can use highlight repair to repair the aberrant color information, or switch to Legacy mode for the Style parameter.

Hope that helps!

[cytan299]

Thank you for uploading that.

This dataset is the absolute worst case scenario for SV Decon.

The noise correlation is extreme; noise clumps/grain are almost as big as the stars;

2023-05-24_14-34.jpg

Did you dither and use bias and/or dark frames? Noise grain should ideally be single pixel in size...

Yes, I used dither, bias and dark frames. Everything was done by the "book". It could be that the STF8300C is just very noisy. I did cool it to -15 degC though.

Nevertheless, SV Decon is still reliably identifying possible star samples (yay - I'm so glad I got that sorted now!).

The next problem with thsi dataset, is that the PSF will be... weird. This dataset is the result of multiple nights.
This has caused data to be stacked that varies wildly in PSF (due to seeing and transparency varying). Indeed, this can be seen around the brighter stars where cores from various nights will either over expose (good seeing) or be smeared out (bad seeing). To minimize this problem, I imported the datasets as L, RGB (e.g. I did not use RGB as synthetic luminance), also because combining data from different cameras for the purpose of luminance is never a good idea.

Thanks for the hint about using L,RGB. I'll try that out. As for the RGB from the ASI2600MC, the problem was that I was actually imaging through smoke from Canadian forest fires. Although they sky looked blue throughout the day, and I wasn't able to see the smoke with my eyeballs, PHD2 kept losing the guide stars during the imaging session.

Despite all these challenges, SV Decon does its best. I did not experience any ringing, though I had to set Deringing Detect to 11%.
Selection_788.jpg
Selection_789.jpg

Unrelated, but do check your color balance (for example using MaxRGB mode), as it appears the image you published is rather green. That appears to be because the star highlights contain aberrant color information, likely cause by the mechanism described above (stacking data with different PSFs). This in turn causes the Color module to come up with an incorrect default color balance. Flipping to MaxRGB will readily show the galaxy being green dominant, which is not correct. Reduce the green bias until it no longer is green dominant, and flip back to "Normal" mode. You should now see more familiar galaxy colors.

You can use highlight repair to repair the aberrant color information, or switch to Legacy mode for the Style parameter.

Hope that helps!

I've always had this "green" problem. For this image, I've flipped between MaxRGB and normal mode but I've found that reducing green too much makes the image look very red. If you look at my log files, I did reduce green but not by much.

Do you mind posting your log, so that I can reproduce what you did? And did you use v1.9 for the processing?

Thanks for your help

cytan

[admin]

Thank you Cytan,

Yes, I used dither, bias and dark frames. Everything was done by the "book". It could be that the STF8300C is just very noisy. I did cool it to -15 degC though.

Hmmm.... multi-pixel co-located defects like these should not be happening in that case;

One thing I haven't tried actually, i using the Correlation Filtering in Wipe. It may or may not help a little.

2023-05-24_14-34_marked.jpg (182.02 KiB) Viewed 29246 times

Thanks for the hint about using L,RGB. I'll try that out. As for the RGB from the ASI2600MC, the problem was that I was actually imaging through smoke from Canadian forest fires. Although they sky looked blue throughout the day, and I wasn't able to see the smoke with my eyeballs, PHD2 kept losing the guide stars during the imaging session.

That explains a good bit. Poor transparency can indeed be troublesome, but especially troublesome when combining such data with other data with better/good transparency.

I've always had this "green" problem.

It should also be said that it is hard/impossible to achieve a perfect color balance with your particular instrument if no luminance filter is in place - as you can see the channel response for both blue and - especially - green increases in the infrared;

(from http://zwoasi.com/product-detail/asi260 ... 8%E6%9C%BA)

Do you mind posting your log, so that I can reproduce what you did?[/quote]

It was a simple processing. The L, RGB loading will significantly change noise considerations in terms of how much you would want to stretch.

Log as follows;

Code: Select all

--- Compose
Parameter [Luminance, Color] set to [L, RGB]
Parameter [Scene] set to [Widefield Galaxy]
--- Crop
Parameter [X1] set to [455 px]
Parameter [Y1] set to [250 px]
Parameter [X2] set to [2978 px (-407)]
Parameter [Y2] set to [1973 px (-237)]
Image size is 2523 x 1723
--- Wipe
Uncalibrated 1 preset
Parameter [Dark Anomaly Filter] set to [6 px]
--- RoI Optimized Develop
Parameter [Ignore Fine Detail] set to [9.8 px]
Parameter [Outside RoI Influence] set to [0 %]
Parameter [RoI X1] set to [1010 px]
Parameter [RoI Y1] set to [756 px]
Parameter [RoI X2] set to [1437 px (-1086)]
Parameter [RoI Y2] set to [1004 px (-719)]
Parameter [Shadow Linearity] set to [12 %]
--- HDR
Reveal preset
Parameter [Context Size] set to [17x17 pixels (0.67% image W, 0.99% image H)]
--- Spatially Variant PSF Deconvolution
Picked some "good" quality samples (e.g. green cores)
Parameter [PSF Resampling] set to [Intra-Iteration + Centroid Tracking]
Parameter [Deringing Detect] set to [11 %]
Parameter [Sampled Iterations] set to [16x]
--- Color
Tricky as mentioned.
Parameter [Bias Slider Mode] set to [Sliders Reduce Color Bias]
Parameter [Style] set to [Scientific (Color Constancy)]
Parameter [LRGB Emulation] set to [RGB Ratio, CIELab Luminance Retention]
Parameter [Dark Saturation] set to [3.8]
Parameter [Bright Saturation] set to [Full]
Parameter [Saturation Amount] set to [200 %]
Parameter [Blue Reduce Bias] set to [1.00]
Parameter [Green Reduce Bias] set to [1.73]
Parameter [Red Reduce Bias] set to [1.18]
Parameter [Cap Green] set to [100 %]
Parameter [Highlight Repair] set to [5 px]
--- Unified De-Noise
Parameter [Grain Size] set to [9.0 px]
--- Unified De-Noise
Parameter [Scale 1] set to [99 %]
Parameter [Scale Correlation] set to [50 %]
Parameter [Color Detail Loss] set to [50 %]
Parameter [Grain Dispersion] set to [9.0 px]

You'd get something like this;

NewComposite.jpg (251.67 KiB) Viewed 29246 times

Color-wise, it still compares reasonably well with other StarTools M101 renditions. E.g. this quick processing of Jim Misti's M63 LRGB dataset;

NewComposite.jpg (75.04 KiB) Viewed 29246 times

And did you use v1.9 for the processing?

Indeed I used the latest 1.9 alpha.

Hope this helps

[cytan299]

Hi Ivo,

Thanks for the log file. I'm learning a lot from you.

As for the lum filter for the ASI2600MC, it supposedly has a UV-IR window already. But I can easily add a UV-IR cut filter to see whether it'll help with the colours.

Thanks again.

cytan

[admin]

As for the lum filter for the ASI2600MC, it supposedly has a UV-IR window already.

Ah, okay. I have just found that in the manual as well. I'm not sure why the spectral response on the website shows clear QE responses in the IR part of the spectrum? Perhaps it's with the window removed.

2023-05-25_12-59.png (103.78 KiB) Viewed 29238 times

[dx_ron]

As for the lum filter for the ASI2600MC, it supposedly has a UV-IR window already.

Ah, okay. I have just found that in the manual as well. I'm not sure why the spectral response on the website shows clear QE responses in the IR part of the spectrum? Perhaps it's with the window removed.

I think the 'no-AR-window' plot is straight from Sony, so it's just the bare sensor.