Startools V1.7 and V1.8 Comparison

[Startrek]

Since downloading Startools V1.8 ,the weather has been unkind for AP, however I was lucky to capture 1.8 hrs of data on M16 Eagle Nebula a week or so ago

As this was not enough data to produce a decent final image I decided to do a general comparison between Startools V1.7 and V1.8 using similar settings

Bortle 8 Skies
95% waning moon on the horizon
6” f6 GSO Newt
ZWO2600MC cooled to -10C Gain 100
L Extreme OSC Dualband filter 7nm
35 x 3 min dithered guided subs
40 x Flats
60 x Bias
Stacked in DSS
Processed in both ST 1.7 and 1.8 but saved as Synthetic Luminance to enable a fair judgment between images

AutoDev
Bin 55%
Crop Edges
Wipe Narrowband preset
AutoDev, ignore fine detail 4.5 pixels
HDR
Sharpen
Decon / SV Decon PSF sampling
Superstructure Isolate and Dim small
Noise Reduction 5.0 pixels
Second Crop
Save as Synthetic Luminance jpeg


Some take always from the comparisons

V1.7
HDR tends to slightly blow out the stars even at default setting of 1.2
Decon is easier to control the level or aggressiveness and easier to avoid ringing around the stars by adjusting mask. Stars do not shrink, tighten or brighten

V1.8
HDR is applied to main structure only and does not blow out stars. Better result but less fine control of range
SV Decon PSF sampling method, has less control of level of Decon effect even using less iterations, stars end up tighter , but some artifacts of pixels in cores of medium to larger stars and also much brighter stars as if they are totally saturated

Both versions have positive and negative outcomes
I tend to agree that V1.8 does provide a slightly better image overall from this limited data but I still can’t avoid some ringing around stars and artifacts in cores of stars.Also adjusting the level of Decon effect is limited on V1.8 , where as V1.7 you can apply any level you wish

Note: This is only a general comparison and based on limited data too

Comments most welcome

Thanks

[admin]

Hi Martin,

It really all comes down to choosing to avail of the extra/improved features of 1.8.

It appears to me the first image is the 1.8 image due to its better handling of noise grain in the darker parts of the image and the improved stars. In particular it shows clear separation of some close-together stars (a hallmark of effective deconvolution). That said, it appears to me deconvolution was applied rather conservatively in the 1.8 image, with M16 showing less detail than in the 1.7 version. Depending on the quality of your data, there may also be scope for correction of the star shapes.

There are admittedly some things you say I don't quite understand (unless you have the versions swapped maybe?), e.g. " less fine control of range " for the HDR module for 1.8 (do you mean 1.7?) and "adjusting the level of Decon effect is limited on V1.8" (again, do you mean 1.7?).

Clear skies!

[Startrek]

Hi Ivo,
Thanks for taking the time to reply
Much appreciated
Firstly I should have labeled the images ( yes the first or top image is V1.8 as you can tell because the stars are really tight and the overall image has more depth )

I have no problem with HDR in V1.8 as I’m getting good results plus gaining a better understanding as well
It’s SV Decon I have a few problems with ( only the Stars shape, odd darker pixels and shadow ringing , not mitigating atmospheric blurring which works really well )
My Stars particularly medium to larger stars in brighter nebulosity regions appear to have excessive ringing or a shadow ring and the cores appear to have a few darker pixels which is not an issue at normal view but appear when you zoom in ( I process both zoomed in and zoomed out to normal view ) Folk tend to pixel peep when commenting on your imaging rather than observe and judge at normal view
Stars also lose their smooth circular format and end up brighter and square when zoomed in , obviously a price you have to pay for Decon to work ?

I’ll go through my workflow for SV Decon on M16

Select SV Decon
Generate Apod Mask Manually
Mask Auto Do
Do not invert , leave stars green
Keep
Press Sampling
Select single round Stars ( 8 to 10 ) with a few green pixel in centre core over all regions of image ( avoid nebulosity areas ) Try to centre Star when selecting
Bump up PSF sample area to 17x17 to ensure blue square just covers white outline and green pixels of sample Star
Press Result
The resultant Stars have morphed into my description above, so I reduce Sampled iterations from 10x to 6x which does help reduce the shadow ring affect

I guess I’m focused too much on the Stars and not on the main structures and finer detail to which Decon does reduce that atmospheric blur

One observation Ive made, due to being oversampled 0.80 arc sec per pixel , if I increase Binning to say 60% , not only does this reduce the noise grain at bit but SV Decon seems to yield better results

Maybe I’m expecting a perfect world ( Stars ) which is impossible

Your thoughts

Clear Skies
Martin

[admin]

I have no problem with HDR in V1.8 as I’m getting good results plus gaining a better understanding as well

Understood. It's the "less fine control of range" that threw me - as long as all is good!

It’s SV Decon I have a few problems with ( only the Stars shape, odd darker pixels and shadow ringing , not mitigating atmospheric blurring which works really well )
My Stars particularly medium to larger stars in brighter nebulosity regions appear to have excessive ringing or a shadow ring and the cores appear to have a few darker pixels which is not an issue at normal view but appear when you zoom in ( I process both zoomed in and zoomed out to normal view ) Folk tend to pixel peep when commenting on your imaging rather than observe and judge at normal view
Stars also lose their smooth circular format and end up brighter and square when zoomed in , obviously a price you have to pay for Decon to work ?

As mentioned elsewhere, stars appearing brighter is to be expected as the erroneously spread energy gets re-concentrated into a smaller area. What also may happen is that any tracking error will become more apparent *if* only atmospheric turbulence is being compensated for. Indeed, the squarish appearance gets more pronounced in the 1.8 image, because indeed, tracking error is present in the image. SV Decon can, however, correct for more than just atmospheric turbulence; by sampling the stars the tracking error gets taken into account, though the degree within which it can correct the tracking error will vary depending on the quality of the dataset. It will never be 100% perfect however.

I’ll go through my workflow for SV Decon on M16

Select SV Decon
Generate Apod Mask Manually
Mask Auto Do
Do not invert , leave stars green
Keep

Be careful here, as it is imperative that the full stellar profiles are present in the apodization mask. A good rule of thumb is that the stars must be totally "obscured" by the mask and stars may not "leak" any of their energy beyond the mask. Mask->Auto->Do will not generate a mask that is sufficient. You may/will need to "Grow" the mask to fully include the stellar profiles. Failure to include/sample the full stellar profile will result in ringing that is more severe than it needs to be (if at all visible).

Once you are roughly satisfied with the result, you will want to activate "PSF Resampling" for further quality enhancement and ringing reduction. Likely "Intra-Iteration + Centroid Tracking Linear" will yield the best results. This feature is rather CPU/GPU intensive and does not really work that well with previews, but is usually well worth it. This feature resamples the locations you gave it, allowing it to actively correct itself for any errors that crept in after each iteration. Ringing will be reduced here, because any ringing that was introduced after an individual iteration will become part of the "new" sample, and thus also gets correct for.

I guess I’m focused too much on the Stars and not on the main structures and finer detail to which Decon does reduce that atmospheric blur

It is totally your prerogative to prioritise one over the other. Though ideally, you would not have to do this at all!

One observation Ive made, due to being oversampled 0.80 arc sec per pixel , if I increase Binning to say 60% , not only does this reduce the noise grain at bit but SV Decon seems to yield better results

This is absolutely true. Deconvolution thrives exponentially on a good signal.

Maybe I’m expecting a perfect world ( Stars ) which is impossible

Deconvolution is not a silver bullet, but Spatially Variant Deconvolution should allow you to greatly improve your images compared to legacy/simpler (non-spatially variant) implementations.

Lastly, if you'd like us/me to have a look at any sort of dataset, please feel free to share one.

Clear skies!

[Startrek]

Hi Ivo,
Thanks for the detailed reply
All great advice
I’ll keep working with SV Decon on some previous data sets as cloudy nights are ongoing at the moment
As always thanks for your assistance
Greatly Appreciated

Clear Skies
Martin

[decay]

Hi Ivo,

As mentioned elsewhere, stars appearing brighter is to be expected as the erroneously spread energy gets re-concentrated into a smaller area.

I assume, in this case the dynamic range of the image (which is previously settled by the final stretch) is adjusted to take this 'brighter' stars into account? So, the stars do not get blown out, right?

Thanks & best regards, Dietmar.

[admin]

Hi Ivo,

As mentioned elsewhere, stars appearing brighter is to be expected as the erroneously spread energy gets re-concentrated into a smaller area.

I assume, in this case the dynamic range of the image (which is previously settled by the final stretch) is adjusted to take this 'brighter' stars into account? So, the stars do not get blown out, right?

Thanks & best regards, Dietmar.

You are spot-on Dietmar in that it may be desirable to extend the dynamic range - from the docs;

Dynamic Range Extension

The 'Dynamic Range Extension' parameter provides any reconstructed highlights with 'room' to show their detail, rather than clipping them against the white point of the input image. Use this parameter if significant latent detail is recovered that requires more dynamic range to be fully appreciated. Lunar datasets can often benefit from an extended dynamic range allocation.

The only problem is that for areas/stars that were already overexposing, it is impossible to know what the right dynamic range is. So the result for those stars may not be correct...

[decay]

it may be desirable to extend the dynamic range - from the docs

Oh yes, I remember reading that. And I remember not understanding it. But now it makes sense, thank you, Ivo. I'll have to look more closely at whether stars are overexposed, because SV Devon often has a huge impact ( ) on my images.