Startools V1.8 Spatially Variant PSF Deconvolution

[Startrek]

Startools Spatially Variant PSF Deconvolution
With all the recent media exposure on Astro forums and Astro channels about BlurXterminator over the past month or so , I decided to post some of my images using one of the excellent modules in Startools, Spatially Variant PSF Deconvolution or SV Decon for short.
See link to Startools Spatially Variant Deconvolution module for an insight into this unique powerful Deconvolution tool.

Ive presented 3 different data sets ( celestial objects) from 2 different imaging sites and showed Before and After images using SV Decon
All images presented in Synthetic Luminance rather than Color to allow easier comparative analysis.

M16 Eagle Nebula
Bortle 7/8 skies
Average seeing conditions
6” f6 GSO Newtonian Reflector
Image Scale 0.86 arc sec per pixel
EQ6-R Pro mount
ZWO2600MC cooled to -10C , Gain 100
Optolong L Extreme OSC filter
2 hours of data ( 150 sec subs )
PHD2 Multistar guiding
Dithered every sub
Flats and Bias
Stacked in DSS
Processed in Startools V1.8

NGC 2070 Tarantula Nebula
Bortle 7/8 skies
Average seeing conditions
6” f6 GSO Newtonian Reflector
Image Scale 0.86 arc sec per pixel
EQ6-R Pro mount
ZWO2600MC cooled to -10C , Gain 100
Antlia ALPT OSC filter
6 hours of data ( 180sec subs )
PHD2 Multistar guiding
Dithered every sub
Flats and Bias
Stacked in DSS
Processed in Startools V1.8

NGC 2359 Thor’s Helmet Nebula
Bortle 3/4 skies
Poor to Average seeing conditions
10” f5 Carbon Fibre Newtonian
Reflector
Image Scale 0.62 arc sec per pixel
EQ8-R Pro mount
ZWO2600MC cooled to -10C , Gain 100
Antlia ALPT OSC filter
4 hours of data ( 180sec subs )
PHD2 Multistar guiding
Dithered every sub
Flats and Bias
Stacked in DSS
Processed in Startools V1.8

Each image was processed in Startools v1.8 as default Synthetic Luminance using mainly default settings in each module ( post final Auto stretch) with some minor tweaks.

SV Decon also was processed generally under default settings ( 10 to 12 x iterations) using between 12 and 16 sample Stars across each image which is very conservative.This module is incredibly powerful.


Auto Stretch for inspection of data
Bin 71%
Crop
Wipe ( Narrowband preset )
Final Auto Stretch
Contrast
HDR
Sharpen
SV Decon ( After only )
Final Noise reduction
Save as jpeg

I chose lower quality data sets to demonstrate that SV Decon works extremely well with all types of data. Obviously higher quality low noise data will produce improved outcomes with SV Decon.

Startools version 1.9 is currently being developed and tested with some exciting new and improved features for SV Decon.

Comments welcome

Clear Skies

[Startrek]

NGC 2359 Before and After images

[Startrek]

NGC 2359 Before and After images

[Startrek]

NGC 2070 Before and After images

[admin]

These are some excellent examples of what real deconvolution can do, and why it is unnecessary to deep-fake detail.

[Startrek]

Hi Ivo,
Thanks, yes the results speak for themselves
No matter how hard you try with alternatives , true high fidelity signal processing will always shine.

Clear Skies
Martin

[Topographic]

Hi Ivo,
Thanks, yes the results speak for themselves
No matter how hard you try with alternatives , true high fidelity signal processing will always shine.

Clear Skies
Martin

Martin

Great job!

A nice reminder.

David

[dx_ron]

Well done!

Are you posting these examples in other places? I hope you point out that ST's decon is actual parameterized L-R deconvolution. It can be reversed, for example (I hope Ivo will correct me if I am wrong about the reversal part)

[hixx]

Hi Martin,
Your images are great!
If I am not mistaken the reversal part is valid as long as You are processing, as You can use the Restore button to revert back to Linear/Original, Wiped or Stretched status. Once you have saved the processed image, it will probably be hard to reverse Decon, but why would one want to do that?
Clear Skies,
Jochen

[Mike in Rancho]

Nice examples, Martin!

Jochen, I'm not sure that's the reversal that's contemplated.

There are any number of ways to undo or restore, if a program allows that return to a prior state. Here, however, I think we are talking about a mathematical reversal. So, a deconvolution PSF modelling could be "undone," probably even iteration by iteration, and it gets you back to the original data.

With the AI box that we presume we have here, altering data based on some kind of external image training - which itself will be continually changing/learning - that may not work.

That said, it seems to me it might be a lot stickier than that. Conceivably BXT could save out log data relating to the alterations made during a particular use, which may or may not permit reversal. But also even with ST, I'm not sure you could start with a finished image and just reverse SVD? You would still need log info, removal of cosmetic repairs like deringing, and probably a construction of the variant PSF that you might need the original linear data for anyway? No?