Experiments in processing lunar images. SC the winner?

[timh]

There are so many alternative ways of processing and sharpening lunar images -- different deconvolution algorithms, alternative wavelet schemes and programs - let alone different parameters and values to optimise that it is effectively impossible to know whether any given chosen path was really the best.

Even beyond that it is anyway only a matter of opinion what is meant by 'the best'. At some point - improved resolution - can just start to look 'over-processed' and of course aesthetic considerations are usually more important -- but less well defined.

So I thought that would try some experiments alternatively processing just one dataset --a .SER capture of Plato -- which crater provides a way to quantify and compare the resolution of alternative images in terms of the number of different sized craterlets that are discernible within it.

The starting supposition was that because wavelet selection and deconvolution are two completely different methods that some mixture of the two might provide a better end result? So - broadly - I compared

a) Autostakkert (15% best frames) + Registax (using gaussian wavelets)
b) Autostakkert (15% best frames) + Registax (using gaussian wavelets)
c) Autostakkert (15% best frames) + Pixinsight (LR deconvolution or BlurExterminator) +/- Registax or +/- Pixinsight MMT (gaussian wavelets)
d) Sharpcap4.1 wavelets (15% frame selection and stacking 279)

To cut a long story short -- surprisingly to me anyway -- the quickest and simplest processing within Sharpcap supplied the highest resolution image in terms of the number of craterlets that could be discerned.

The Autostakkert routes provided good results -- and deconvolution provided good sharpening that was somewhat enhanced if preceded by a wavelet process but at least my hands and in terms of resolution Sharpcap won. Of course there are so many variables and it could well be that a more skilled user of LR deconvolution ,Registax, MMT could have done better but is was striking to me that SC worked so well and with such relative speed and simplicity.

Maybe the key is in the particular 'contrast' frame selection algorithm that SC is using in comparison with the Autostakkert selection algorithm?

The other thing that I found remarkable is that - apparently -- the resolution could so approach the theoretical maximum for a 12 inch telescope on what seemed a fairly average night that I am left wondering whether any improvement is possible even -- could stiller skies even have made much of a difference?

Tim

Images are from process d) Sharpcap alone and process c) using LR pixinsight deconvolution and MMT followed by a final step of Registax wavelets. Both images were also adjusted in Affinity for contrast, brightness etc. They ended up looking overall rather different in terms of final adjustments but the SC only processed appears to more clearly show most all of the craterlets -- but in some ways the other picture may be better? Any comments welcome --- undoubtedly I may have made mistakes in getting the best out of the images.




-------------------------------------------------------------------------------------------------
Capture of Plato SER file in Sharpcap was as described below viewtopic.php?t=8608

with the following details..

Output Format=SER file (*.ser)
Binning=1
Capture Area=1600x1200
Colour Space=RAW8
Gain=89
Exposure=16.7380ms
Timestamp Frames=Off
White Bal (B)=75
White Bal (R)=55
Brightness=15
Duration=105.075s
FrameCount=1862
ActualFrameRate=17.7207fps
TimeZone=+1.00

[timh]

Just to clarify. Here is a bigger image of just the crater. In some ways the image subject to more complex processing looks better but the SC only processed image shows the craterlets more clearly

[Jean-Francois]

Hello Tim,

Do I understand correctly ... you take a film with SharpCap and then you process it again in SharpCap or the other software.
So that all your processing used the same data ...?

What is the difference between a) und b) ?
Do you use the Multi Point from Autostakkert ?

Your comparisons have 2 components ... first how one software perform the stacking, second how it perform the detail enhancement.
A lot of people are still using Registax for the Wavelet processing.
For the stacking ... you have the PSS (or the old but still giving good result Avistack software).

Best regards,
Jean-Francois

And yes ... if you can wait long enough, you can achieve some times on some field the theoretical resolution of the optic.
At least you can narrow it. Only with the assumption that the optic is good manufactured and collimated.

[Menno555]

Thanks for sharing Tim!
Not doing that much lunar imaging here, but these kind of write ups open new things to try
Btw, maybe here 2 other apps to try with lunar stacking:
PlanetarySystemStacker. That app gives a bit sharper images in comparison to Autostakkert after stacking.
ImPPG (Image Post-Processor). This stand alone app is purely for deconvolution, unsharp masking, brightness normalization and tone curve adjustment.

Menno

[admin]

Hi,

an interesting post, thanks for sharing

As Jean-Francois says, separting out the stacking and sharpening would be an interesting next step - SharpCap is somewhat limited in stacking because it works by analysing images on the fly rather than looking at all the images in the video before deciding on the best ones to select. However, the sharpenng in SharpCap may give a bit more control that Registax due to the larger number of wavelet layers that can be adjusted (particularly in 'Expert' mode in the latest builds.

cheers,

Robin

[turfpit]

Tim

A tricky project trying to evaluate the different ways to process a lunar image. Deconvolution is helping, here are a couple of links:

https://www.cloudynights.com/topic/8471 ... vid-biggs/
I watched the video in the first post. I never realised that deconvolution has an application in medical imaging. Some decent images came out of that workflow.

https://chaoticnebula.com/pixinsight-lu ... nvolution/
Here it looks like trying different settings and eyeballing the results seemed to work.

Try experimenting with lower settings in Registax, see viewtopic.php?t=5450 . I found (over time) that layer 2 had the most effect (at least on my equipment) and learned that with sharpening 'less is more'. I never use layers 4, 5, 6 now. I have found it is always worth trying the AS sharpened (_conv) file in Registax as it sometimes gives a better result.

The atmosphere is always a problem - impossible to predict. For a lunar session, I now use IR685, UV/IR Cut and Wratten#25 (or red) filters and gather data with each filter. In the Clavius session I collected 30Gb of data in just over 30 minutes. That is another strategy for beating the seeing. Using Virtual Moon Atlas to plan lunar imaging sessions has made me more productive at the scope.

Dave

[turfpit]

Menno

Thanks for the 2 links. I will have a look at these on one of those rare cloudy nights. I did an Astrobin search for imppg, it looks to be a tool favoured by solar imagers.

https://app.astrobin.com/search?p=eJy72 ... AKJ0K3M%3D

Dave

[timh]

Many thanks indeed everyone for all of those useful responses and points. Still experimenting with this – will try out some of the software suggestions - Thus far SC still apparently yielding the best resolution.

A few specific replies

Jean-Francois

Yes you are correct. Here I took just one .ser file – just under 2 min comprising 1863 frames --and experimented with different methods of selecting, stacking and processing it.

It was a particularly useful dataset to start from because – fortuitously – it caught a very low sun over Plato which rendered the craterlets within it visible. These are of ~ known sizes varying from 3 down to about 1 km making it easy to compare resolution cf https://www.cloudynights.com/topic/3484 ... raterlets/ and below
You asked about how the stacking and processing was done within Sharpcap. For this I used the folder camera, dropped the .ser file into SC4.1 , set exposure/ gain (arbitrary) for a reasonable-looking histogram, opened the planetary livestack tab, set it to multipoint, surface alignment, align RGB, set frame selection to 15% and – accordingly - the size of the stack to 280 and finally adjusted wavelets (autosharpen more) and brightness/contrast to default values . I let the ser file cycle round and saved a TIFF file when the contrast was highest and the detail looked good. A very simple and quick process for getting a good image!

a) and b) are the same. Just an editing mistake

Yes I did use surface stabilization and multipoint in Autostakkert
Yes you are right about the need to separate out the different elements – the quality selection, the stacking and the detail enhancement aspects of processing.

As far as I am aware the VX12 optics are pretty good – or at least I can say that the collimation using the Howie Glatter and tublug looked accurate. The apparent resolution of close to 0.6 arcsec or so is consistent with that.

Robin

Thanks for your comments (and the software!). Yes it could very well be that the crucial difference lies in the SC wavelet software –maybe also that it is quicker and easier to get to settings that are at least almost right with the autosharp shortcut buttons?

I did also wonder though –maybe particularly in this case – whether the continually updated nature of the current selection of the best (15%) stack could actually be a strength rather than a weakness ? Selecting the ‘best’ frames that are temporally more or less contiguous with each other might be better than selecting them from across the entire 2 minute period ? Some of the craterlets have very low sides and the shadows that make them more visible possibly only apparent for tens of seconds within the 2 min window?
Anyway -whatever the reason - based purely on craterlet resolution the SC output still appears the best – with just minor further improvement possible using deconvolution in PixInsight.

Menno,

Thanks for these links. PlanetarysystemStacker certainly looks interesting to try

Dave,

It is indeed a tricky and complex area and the simplest explanation for my relative failure with the non SC software is that I am just not doing it right in some way (though I did use the .conv file and tx for the registax tips).

Of course –very happy to share the .ser file with you or anyone else that would be willing to have a go with it. Prizes for the best resulting image ?

A devil’s advocate comment though on the importance of finding better atmospheric conditions. What puzzles me is that I can measure the final image resolution to have been down to at least ~ 0.7 arcsec which is pretty close to a theoretical best of 0.5 arcsec which surely could not be much exceeded whatever the conditions? Does this show that you can in fact get really good resolution even from not particularly good data sets (see the Autostakkert quality graph below) simply through the power of having enough frames and applying selection? But I recognize of course that the ‘quality’ scale is not linear and in all probability an actual 0.5 arcsec image from a still night would look considerably better than a 0.7 arcsec? Anyway I’d love to have the data to make the comparison.

Tim

[turfpit]

Tim

Some good detail there on the craterlets. I have managed to capture 3 using the Celestron C8. Now where did I put my LRO?

I would be willing to have a go at processing the data from the VX12. How big is the stack from Autostakkert?

Dave

[Leandro Salgueiro]

Excellent instructive work, very good details
Congratulations.
Leandro