Astro to Astro-art Continuum, 2024 Edition

[Mike in Rancho]

Porting this over to stop my derailing of Freddy's Tadpoles in the gallery forum, here (sorry, Freddy!): viewtopic.php?f=11&t=2933
Thanks also to Dietmar for (of course) finding Ivo's table on this subject from a couple years ago, here: viewtopic.php?p=11393#p11393

The discussion in Freddy's post involved unsharp mask and I suppose sharpening in general, as an astro processing tool.

Since the time of Ivo's table of course, things have moved on (and with the explosion of "AI" are likely to continue), though the basic concepts should remain the same. GHS has perhaps become the stretch of choice across several platforms. Various plugins including the X-tools are rampant. In fact, they seem so heavy-handed (or good at what they do, depending on your view) that I've seen decent-looking images posted along with kudos for "you have great data here," yet when I open the same shared file to inspect it with OptiDev and Wipe, it's actually a complete train wreck. Finally, I thought Topaz was dead in the water, but to my horror saw continuing usage recently. And it's still growing hair on things!

The latest iteration of BXT, and AI in general, have also sparked interesting discussion not only as to the legitimacy of its deconvolution and new "perfect star repair" mode, but also as to perhaps ultra-purism. Of which there seems to be one last devotee, that I can tell.

With the caveat that of course one can do whatever one wishes with their data, processing-wise, though hopefully workflow is explained, and that we all have different goals and lines in the sand we won't cross (usually), I'm wondering where we think things today fit into the table (recreated here), and if it needs to be expanded.

I had forgotten that Ivo lumped Sharp, Contrast, and HDR all together. Does that seem right? Following onto the previous discussion, it struck me that due to use of a blurred subtraction, sharpening is a bit manipulative of reality, even if technically intrinsic. Contrast and HDR seemed more akin to stretching, sometimes locally, and thus affecting local relative pixel intensities, but not actual re-jiggering of data the way a subtraction would in a sharpening tool. True or no?

I also wonder where things like OptiDev, other stretching tools, and Denoise would stand. Or SS, for that matter, since it is an oft-utilized ST module and is permitted in the normal tracking workflow.


2021 Ivo Table

	Restorative		Enhancing
Intrinisc	Deconvolution Color Wipe	HDR Sharp Contrast	Shrink
	DBE (sample setting-based background removal)	Heal (hot pixels, dead pixels)
Extrinsic	Photometric Color Correction		Heal (dust donuts) Neural Hallucination (TopazAI, StarNet) Synth Operations with hand-drawn masks

[decay]

Thanks for porting this over, Mike.

Well, since it was me propagating the discussion about Unsharp Mask, (local) Contrast Enhancement and Wavelet Sharpening, here are my thoughts about the related StarTool’s modules:

Yes, I’d say it is right to put Sharp, Contrast and HDR modules together in one category. As I tried to write down in Freddy’s thread, I assume that they do work in similar ways. Sharp uses an implementation of Wavelet Sharpening and as far as I understand this makes use of blurred subtraction, just like Unsharp Mask, but more sophisticated. This seems to be technical way to implement this local contrast enhancements (or maybe similar manipulations) – and yes I also think that this is a manipulation of reality.

And the Contrast module works just the same, I guess, but at a larger scale. The ‘Locality’ slider is used to dial in the scale at which the local contrast enhancements will work. But nonetheless these are _local_ contrast enhancements and for me _local_ contrast enhancements are always a ‘ manipulation of reality’ (to use your wording). (This is not true for ‘normal’ contrast implementations, since they do a global contrast enhancement, which is just ‘stretching’ in the end.)

And the same goes for the HDR module. It works at smaller scales and at distinct brightness levels, but again it uses local contrast enhancement. (Of course, I don’t know, if there’s also a kind of blurred subtraction at work, but maybe it’s not important, how it is done?)



(Of course, often (or most times?) we do need all this, because of the huge brightness differences in AP!)

Optidev (and other stretching tools) I’d say is more on the ‘restorative’ side of the table.

Best regards, Dietmar.

[Mike in Rancho]

If you see that electric glow around objects, then it is just applied too much, Mike. You can see that if you apply the flux module also...
Unsharp mask can do great things but it is always a balancing act. But i think that counts for just about every processing too. And someimes , it does not add much either, it all very depends.Unsharp mask might (mind the word might) accentuate things in the picture, well if they are there. Eg i am now working on the Christmas tree, taken in bad moonlight, and really , unsharp mask just does not do much there for whatever reason..;

It is thesame with stretching...it is a hard thing to do. Often one thinks it is pretty ok, but after finishing the image, often it is applied either too much , or not enough..it is a tricky thing imho. I can't really do it well in PI in most cases...Startools is way better imho. I could not make an M42 with an descent core, it came out always way overblown in PI. Strangely in Startools , it is not hard to do...or at least let say far more easier.

Processing is a bit like Pandora's box. Often it goes bad, but the hope remains on a next occasion you will do it better and eventually get a descent image

Hi Freddy. Quoted over here to leave your tadpole alone and to maybe go beyond sharpening.

Parameters and settings are important, sure. We also know that from known restorative tools, like deconvolution, where we can end up with artifacts and errors. So it's not really an issue of whether a light touch makes everything okay (people said that about Topaz too), but whether a tool starts bending reality (as defined by the captured data) too much. And then the ensuing questions are - what is "too much," and exactly what reality is being bent. And thus, we may consider some such alterations to be "acceptable." If one even care,s of course.

Yes, I’d say it is right to put Sharp, Contrast and HDR modules together in one category.

You may be correct, Dietmar. Perhaps I am being too harsh on Sharp, or too easy on Contrast and HDR?

I tried to do some further reading, though that all tends to be generalized photo processing, and I still didn't look up acutance. In unsharp mask, they say the blur acts as a high pass filter. Edges are detected by that blurred subtraction, and then added back in. The radius controls the size of the blur. The claim is that structure sizes are not changed, but the transition between intensities is narrowed. Thus a perception of harder edges. Is that creating detail, perhaps where a true transition should be softer, and how would one even know?

Using ST's wavelet Sharp, I haven't really noticed edge narrowing. Instead it seems to brighten brighter areas and darken darker areas. That may be due to leaving things mostly default and not adjusting the wavelets or the initial scale setting. Of course there are even further sliders as well.

I also looked up localized contrast enhancement and indeed, was surprised to learn how close of a relative that is to unsharp mask. It just uses a larger scale blur. I suppose I will have to try that too manually, in Gimp layers, to see if I can figure it out.

I was under the impression that Contrast and HDR were more like local re-stretching than applying image filters.

[dx_ron]

Thanks for pushing the conversation.

The first step is that I'm not clear about is the word "local". Does local mean that HDR, for example, changes the stretch curve so that it is different at different coordinates within the image? That would be akin to a "masked stretch", right? Or, does local mean that ST has analyzed the image and found regions (coordinates) where a lot of the overall dynamic range is concentrated and then it modifies that particular region of a global stretch curve. That would akin to setting the inflection point in manual GHS to some part of the image that is currently at a brightness range where you want to increase contrast.

As for wavelet sharpening - planetary imaging would hardly exist without it. It certainly causes artifacts if pushed too hard, but it seems to be possible to use settings that do not invent detail. But maybe the type of structure detail common in a planetary image is fundamentally different than a dso image?

[decay]

Does local mean that HDR, for example, changes the stretch curve so that it is different at different coordinates within the image? That would be akin to a "masked stretch", right?

Yes and yes, Ron. I'm currently preparing a new post, which may take a while. But hopefully it will make things a little bit clearer.

Dietmar.

[Mike in Rancho]

That would be akin to a "masked stretch", right? Or, does local mean that ST has analyzed the image and found regions (coordinates) where a lot of the overall dynamic range is concentrated and then it modifies that particular region of a global stretch curve. That would akin to setting the inflection point in manual GHS to some part of the image that is currently at a brightness range where you want to increase contrast.

Can't help with GHS. The time I tried to use it, I found it as confounding as PI's HistogramTransformation.

Interesting surmising though. Perhaps the negative connotation to masked stretching is only meant for the main global stretch. And as for localities and local stretching, perhaps that becomes acceptable - if not mandatory to accomplish the desired task - once one has taken a step towards the "enhancing" categories.

I have to go back to the Features and Docs, but I think at least one of them (HDR?) discusses that it is effectively local stretching and local gamma alteration.

Query how it is done. I think some programs accomplish locality by way of tiling. PI's LN stacking may do that, and perhaps BXT. Unsure how tiles are transitioned. I've always assumed that SVD accomplishes locality (spatial variance) by fully variable mathematics based on the sample stars.

That said, I have developed the sense that ST uses temporary masks in the background at times, which we can never get to. Probably based on the stashed away linear data. And so, for HDR that could be a luminance mask? And the various sliders and options control that mask?

There are several modules where it seems clear this kind of thing is happening (i.e. ever seen a dropdown menu with "linear brightness mask darken" in ST?) but we never actually see the actual mask.

Oh well, still guzzling coffee so I may not be fully coherent yet. Waiting on Dietmar's promised dissertation to better understand things.

[dx_ron]

The only thing you need to know about GHS in the context of what I said is that there is a single stretch curve for the whole image. I suppose PI allows you to make masks and have different GHS stretch curves in different places. I've only used GHS in Siril, which has no mask abilities at all. The idea is to use GHS to manually do what Ivo calls "allocate dynamic range". If there's a specific brightness range where you want to highlight small differences ("increase contrast") you can do that with GHS. ST evaluates the overall image to find brightness ranges that are "crowded".

[decay]

In unsharp mask, they say the blur acts as a high pass filter. Edges are detected by that blurred subtraction

Blur itself indeed is a low pass filter, but yes, the subtraction of this blurred result from the original image acts like some kind of high pass filter. So

original image (full bandwidth) – blurred image (low frequencies only) = detected edges (high frequencies only)

I’m not sure, if this ‘blurred subtraction’ is mainly a remnant of analogue film processing. The first thought of course, is to use directly a high pass filter instead of this ‘blurred subtraction’. And this actually seems to be the general principle:
https://archive.nptel.ac.in/content/sto ... /8_32.html

Maybe leaving out the mathematical part, the text and the diagram are quite revealing.

Is that creating detail, perhaps where a true transition should be softer, and how would one even know?

Absolutely right! This way _every_ transition will be sharpened, even if the original structure has a soft transition.

I again would like to point out this Wikipedia article. It’s really worth reading, at least up to ‘local contrast enhancement’:
https://en.wikipedia.org/wiki/Unsharp_masking

And there it's written down:
“The resulting image, although clearer, may be a less accurate representation of the image's subject.”

As for local contrast enhancement and 'masked stretching' (Ron!) I found this article:
https://www.cambridgeincolour.com/tutor ... cement.htm

It explains nicely illustrated, how this local contrast mask is created and how it is used to control which contrast aka ‘stretch’ is applied at which position in the image.

(Edit: The important explanation is the note beneath the second illustration: "The "mask overlay" is when image information from the layer above the local contrast mask passes through and replaces the layer below in a way which is proportional to the brightness in that region of the mask. The upper image does not contribute to the final for regions where the mask is black, while it completely replaces the layer below in regions where the local contrast mask is white.")

I guess, these are the ‘hidden’ masks, you think ST uses behind the scenes, Mike?

Running out of time, so no dissertation Does this help for our discussion, anyway?

Dietmar.

[jlh]

I am certainly not well versed in what is a true astro image vs. what has things added that are not really there. As to Sharp my question is whether it is making the image sharper than the original photons support? Or, is it correcting things that prevent those photons from otherwise being seen and/or presented in their original sharpness? My limited understanding of the unsharp mask is that it was designed to correct the blurriness caused by certain shortcomings in analog photography (perhaps color photographs in particular). By extension is the sharpening algorithm in astrophotography a similar type correction? Of course, like everything, provided it is used in moderation.

Jeff

[Mike in Rancho]

Thanks, Dietmar.

I'll try to dig into all of that more later. Had to do an office day today, and then tonight was clear and somewhat decent seeing (MB all bright green indices though the base seeing is only 1.6" so, meh) so the rig is out catching some more Horsey before days and days of rain and clouds come in starting tomorrow.

That Cambridge site is actually where I read through a few explanations of these generic processing matters, including their local contrast formula!

I guess, these are the ‘hidden’ masks, you think ST uses behind the scenes, Mike?

Actually no, though I did catch onto the way the term "mask" is being used for these techniques, which at their base level just seem like a sequence of Layer (or Pixel Math) operations.

I'll have to go check on the modules, but I was leaning more towards a true mask, albeit one probably made based on the linear luminance.

I am certainly not well versed in what is a true astro image vs. what has things added that are not really there. As to Sharp my question is whether it is making the image sharper than the original photons support? Or, is it correcting things that prevent those photons from otherwise being seen and/or presented in their original sharpness? My limited understanding of the unsharp mask is that it was designed to correct the blurriness caused by certain shortcomings in analog photography (perhaps color photographs in particular). By extension is the sharpening algorithm in astrophotography a similar type correction? Of course, like everything, provided it is used in moderation.

I don't know, Jeff. I guess that's what we are here to discuss. Via the low/high pass filtering, and/or layer operations, however you want to look at it, "edges" are being detected and then hardened by having their transition zones narrowed. Can that be thought of as a correction? Well, it's an enhancement, and I imagine it can "seem" to correct errors due to being out of focus or having motion blur (I used UM on a flying red tail once and it helped).

A true correction would be deconvolution, at least as to those matters the synthetic modeling is designed to undo, or that the PSF sampling has actually measured.