Version 1.8 HDR and SV Decon

[Startrek]

I recently updated to version 1.8 and spent quite a few hours using some recent datasets trying to get HDR and SV Decon to work.
I’ve read the user notes but still not getting satisfactory results on these 2 important modules
HDR tends to be way too aggressive using the presets compared to v1.7.
SV Decon after creating a mask and going through the long winded procedures just dims my stars
Going through the user notes it seems way to complicated and time consuming to process its requirements to achieve a result. I can use Decon in v1.7 and achieve a great result a minute or 2

Unless I can find a suitable timely fix for HDR and SV Decon , I’m staying with v1.7 as it just works for me

HDR , SV Decon , NB Accent and a couple more SHO options in the Color matrix are the only changes I’ve come across , otherwise it’s almost on par with v1.7

Cheers
Martin

[hixx]

Hi Startrek,
V 1.8 is adding a lot over 1.7 version. Here`s an Overview over all new features in V1.8:

• The 'Spatially Variant Decon' module is new to Star Tools 8. Its unrivaled functionality allows optimising the Point Spread Function locally, based upon multiple star samples.

• The brand new 'Narrowband Accent' module is providing an overeasy way to add narrowband accents to visual spectrum datasets. A seperate narrowband processing chain has been added to Luminance and Color paths in all modules, allowing indepenend but simultaneous processing.

• 'Repair' module has been made tracking aware, it may now be included in the workflow between 'Color' and 'Denoise'

• 'HDR' module is also tracking aware now, featuring a number of presets for common use-cases.

• A 'Correlation Noise Filter' in 'Wipe' module was added to remove any correlated effects such as debayer artifacts, so any downstream module may better focus on detail due to higher SNR.

• The new 'Protection' parameter plus enhanced deringing in 'Sharp' module will make it less prone to singularities, hence reducing artifacts.

• Further improvements include improved Tracking, improved Denoise and various code optimisations

• Native support for Apple M1 ARM CPUs.

• 'Layer' module now offers 2 new Filter mode options: 'Gaussian Highpass filter' and 'Laplacian over Gaussian 0 crossing' plus a new Cap mode 'Add 1/2 unity + clip'

• New 'AltStars' preset in AutoMask generator uses a star detection algorithm that is much better at detecting stars with full stellar profile included.

In general Your observations seem to be quite strange, but seem to result from using ST8s parameters as You would use ST7's Understanding ST8s parameters should help You yield results closer to Your expectations.
SVDecon should not dim stars, but should provide clearer detail throughout the whole image. There are settings though, which might cause stars to shrink to a sharper and less "blobby" look. (which is part of Deconvolution reversing the Athmospheric seeing. E.g. Should You process Hubble images, there is little need for Decon at all. You may control the strength to taste using the iterations and radius parameters. In fact, ST8's SVDecon may operate pretty similar to ST/'s Decon, depending on settings.
Same is for Contrast and HDR: You are not bound to the V1.8 presets, there are hints in both Guy's User Notes and the Inofficial Manual how to mimic "V1.7 - old style" HDR presets. Please also note, in HDR You may tune each Detail boost and Gamma values for both Highlights and Shadows individually
Bottom line: ST8 provides all of ST7, but also gives You more options and features.
Adding more specifics on Your settings as well as the nature of the subject should help in providing more detailed advice.

Cheers,
Jochen

[Startrek]

Hi Jochen,
Thanks for your reply
I’ve spent more time on HDR today and seem to be getting better results when using Optimise and Tame
presets and the sliders below
Still working on SV Decon ….. will report back later on
Just to advise I image with a OSC 2600MC not Mono so NB Accent is of little use to my imaging and processing ( I just use dual band OSC filters L Extreme and Antlia Golden )
I don’t use and never had to use Repair , Lens , Heal etc…
I’ve been a Startools user since 1.4 and maybe my intro and brief experience with 1.8 was a bit unfair and irresponsible so take it with a grain of salt.
Startools is obviously developing into a far more sophisticated processing tool which obviously will need more time and experience to master

Clear Skies
Martin

[Stefan B]

Hi Martin,

Just to advise I image with a OSC 2600MC not Mono so NB Accent is of little use to my imaging and processing ( I just use dual band OSC filters L Extreme and Antlia Golden )

If you never shoot unfiltered or with a broadband filter it's really of little use for you. But with a OSC and dual NB filters, NBAccent can be a great tool. I frequently use it for combining broadband and duo NB data to get natural star colors and nebula details alike, see here e.g.:

https://www.astrobin.com/w3po4i/C/
https://www.astrobin.com/2pahbr/

So NBAccent isn't necessarily for mono cams.

Regards
Stefan

[admin]

Hi Martin,

FWIW, it is entirely possible to "neuter" version 8 and closely approximate what you were used to in version 7 (including processing times), though you would really be missing out in terms of enhanced noise mitigation, improved detail recovery and enhanced dynamic range control;

Untitled.jpg (352.36 KiB) Viewed 15242 times

(set Signal Flow to "As Visible" and even the noise grain will be the same).

It is possible the HDR presets may appear to be a bit more dramatic, but they - like other modules - now tend to yield predictable results no matter the dataset, while quickly demonstrating the difference between the settings. Using presets only to process your images is, of course, not typically recommended as it relegates any artistic vision you may have for your dataset to that of the programmer. Presets are meant to be good starting points for common scenarios.

The SVDecon module is a pretty big deal, as it is the first (that I know) of deconvolution solution that can take slight (or big) changes in the Point Spread Function across the image into account. You should be noticing dramatic improvements in detail recovery over the version 7 algorithm . Stars definitely should not be dimming (their energy should be conserved, just relocated/re-concentrated) and I am really wondering what might be going wrong here! Stars may "lose" energy due to clipping if the energy concentrated makes the star "whiter than white" of course; they should appear "tighter".

But even if you don't want to use star samples at all, you can still avail of some slight improvements over the version 7 algorithm (more decon of highlights, sub-pixel decon). 1:1 comparison of 7 vs 8 is a little hard from within the two modules, as 8 retains more of specific type of noise grain for final noise reduction to take care of/shape. 7 would remove this during regularization, typically making things look blurrier (the old Error Diffusion parameter would somewhat thwart this, but not all of it).

If you have any specific images or datasets you'd like us to look at, do let us know!

[Mike in Rancho]

Hi Martin,

Honestly your first go-throughs are not all that different than it was for the rest of us who followed along through the alphas and betas and MR's of 1.8. I think more than a few of us still kept 1.7 on the side.

But eventually that fades away. I don't even have the shortcut to 1.7 on my desktop anymore. Plus, there are many improvements overall, as well as bug fixes, even beyond the list provided by Jochen, that got implemented throughout (and I don't think were retro'd to 1.7).

Before long the SVD star selection will become second nature and quicker. In almost all cases the result will be much better than 1.7. Sometimes datasets aren't deep enough or didnt' retain enough resolution to get a good SVD, so you can default to synthetic. The difference being the way masking works, but you'll get used to that too. Bump up deringing parameters if you need to, and touch up mask errors if needed. Blown stars sometimes end up not having a full profile captured by the automask. I tend to save centroid tracking to one of my final steps since that increases the calculating time.

HDR, like SVD, depends on the data in front of it. But if the effect seems too harsh, one of the easiest ways to soften the blow is to just lower the tame parameter from the default 1.25, even down to as low as 1.05, if you want softer results.

In a month or two you probably won't be harkening back to 1.7 all that much, or even at all. Just guessing.

[Startrek]

Thanks Stefan, Ivo and Mike ,
Appreciate your replies
I guess I’m guilty of being less tolerant of change in older age but on the other hand change is healthy in your later years
I’m more confident with HDR now so more time and more data sets should put that module to bed sooner than later

It’s SV Decon that I’m struggling with at the moment ( I’ve only been using the Synthetic ( easy ) option but still playing with the Multiple Star sample option

Using the Star sampling option ….
I open SV Decon
Select a Star Mask ( is it necessary to use extra sensitive or general Mask ?? )
Draw a small rectangular area for future inspection
Hit the Sample button (screen turns red with stars circled with dark red lines
I select stars that have green centres on 4 sectors of the image but outside nebulosity areas (say 8 to 10 stars across the whole image ) Usually the larger stars have green points in the centre ( not sure if I can see blue )
Press Result at top
Review SV Decon Test Rectangle area against existing area
Once satisfied with result Press All
I noticed straight away that the whole image has dimmed somewhat, is this normal ?
Also a few of my medium to larger stars have a few pixels in the centre dimmed or darkened which only can be seen whilst zoomed in at 150%. Is this normal

It’s the sliders which I need to get an understanding about

Thanks Martin

[hixx]

Hi Martin,
For SVD You'll need an apodization mask created, which is a bit different from the star mask used in ST7. This might throw You off...
In ST7 the (inverted) Star mask was mainly needed to prevent Decon work on Singularities (bright star centers , creating arifacts. The Apod mask, however is creating an (analytical) star mask to identify potential sample star candidates (on top of that it serves as an indicator for singularities).
Using SVD should be easy as 1,2,3.... It's all about selecting good sample star candidates.
Note: if You work with a preview area, you need to select sample stars within that area.

• 'SV Decon' removes seeing and/or optical diffraction customized to multiple image sections .

• Select preset as per image type as a baseline - Deep Space or Lunar/Planetary.

'Deep Space'
Method
For DSOs:
This is a way of using the module which should give good results in most cases:

• Select 'Auto-generate mask' to create an apodization mask - a form of star mask - or use the ApodMask preset to create one - this masks out overexposed stars which cause ringing.

• Select a preview area rectangle in the image to see the effect of changes - this speeds up analysis. Choose a bright, detailed and noise-free area.

• Click on 'Sampling' button to show the possible sample stars.

• Keeping Zoom at 100% - should allow the stars should be large enough to evaluate and for you to position the sliders at the edge of the image easily to cover the whole image in 6-9 segments.

• Select samples - at least two from each segment

• Samples should be green without red centres. See 'Choosing sample stars' below.

• Add more samples if there is a lot of variation in the way the stars are deformed.

• Put a sample in any area where the stars have been deformed in a slightly different way.

• Adjust 'Sampled PSF Area' so that the blue area just surrounds the selected stars green pixels.

• Increase 'Sampled Iterations' until you see no further improvement

• Set 'Synthetic Iterations' to 'Off' unless you want to use an optical PSF model

• Zoom in and out so you can see the effect in the detail and as a whole.

• Toggle top "Pre Tweak/Post Tweak" button to see effect of last adjustment if needed.

• Normally other default values work well - but you can experiment if you want.

• When done, select 'All' to apply this to the whole image - this may take some time.

• Press 'Keep'.

For Lunar/Planetary targets:
'Lunar/Planetary'

• Click the 'Plnt/Lnr' button

• Don't try and select a sample star

• Select a Synthetic PSF model to model the atmosphere.

• Increase 'Dyn. Range Extension' to increase the dynamic range.

• Increase the 'Synthetic Iterations' until no improvement is seen.

• Adjust the 'Synthetic PSF Radius' until ringing occurs - then back off a little.

• Press 'Keep'.

For the Apod mask, I use the extra sensitive option a lot, but You'll need to try which option works best for You
Hope this helps. Again, for more background & details, I recommend to read the SVD chapter of the Unofficial manual (available in download section)
Cheers,
Jochen

[hixx]

Martin,
Regarding the grey star centers...
This is something we already observed in the ST8 alpha and it is very much improved in the final version. To mitigate this effect, my guess would be to use less iterations and/or try reducing Linearity Cutoff or increasing Dynamic Range extension parameters. As usual with complex algorithms like Decon, there is no silver bullet, but a range of parameters that may tweak the result towards the right direction. Maybe someone else can add detail or other options here.

Good luck,
Jochen

[Startrek]

Jochen,

Thanks for the detailed procedures
Brilliant !!
I didn’t realise you need to ensure you select a sample Star within that preview rectangle area
I could be wrong but I don’t think Guys notes mention that specific requirement
As far as some large stars having a few odd dull pixels , maybe Ivo can correct that issue in ST v1.9 ??

Also I notice that after Decon the stars appear like bright disc and have lost their natural fade from centre to outside ( I suppose you can’t have everything) They are definitely tighter but a bit artificial looking
Is that your opinion of the Stars after Decon
In v1.7 Decon I had more control over the stars outcomes leaving them more natural looking . I just grew the mask and backed off the pixel size a bit
Your thoughts

Thanks again
Matin