The Halos that Aren't

[Russ.Carpenter]

Ivo, I could use your help. I am frequently asked to explain why StarTools processes stars (especially fat stars) differently. In general, astro imagers are used to bloated stars with undifferentiated profiles. They believe that StarTools stars are either a processing error, or an inherent flaw in the software.

In your recent dialogue with Mike in Rancho you said

One of the things I created it for, is to show people why stellar profiles look the way they look; the amount of times you still see people describing diffraction patterns as "halos" is too damn high! They are a natural thing, inherent to optical trains, and they can tell you so much about the optical train itself, as well as provide you with extra information on stars (the rainbow patterns).

I would be very grateful if you could boil down your philosophy to two sentences. I'd like to quote those sentences in my postings of StarTools images, particularly in Astrobin.

Thanks!

Russ

[admin]

Hi Ross,

It's not so much a philosophy - it's just properly respecting the data, as is the job of the astrophotographer. Clipping (throwing away) or hiding signal is never warranted.

Stars are diffracted, single-pixel point lights - they should remain so as much as practicably possible in your final image, overexposed or not. Mostly, this means judicious highlight management after careful deconvolution.

Good highlight and diffraction management is paramount, particularly where non-stellar objects could otherwise be mistaken for stars, for example in this annotated image you posted hours ago;

Blow out stars, and you risk blowing out all the other annotated faint fuzzies.

Obfuscate diffraction patterns, and it becomes much harder (or impossible) to discern what is merely a diffracted single-pixel point light (e.g. a star), and what is fuzzy multi-pixel object (galaxy, planetary nebula, etc.).

Furthermore, being able to see detailed, consistent diffraction patterns in stellar profiles, yields a lot of clues as to the veracity of detail, focus, and characteristics of the optical train.

A poor image hides, and obfuscates these diffraction patterns and important details, either accidentally (by a poorly chosen non-linear stretch) or intentionally (to obfuscate after-the-fact untoward manipulations, such as introduced neurally hallucinated detail).

In short, in StarTools we take these important aspects of AP image processing very seriously - we seek to recover true detail, signal fidelity and truth.

Hope that helps!

[Russ.Carpenter]

Hi Ivo,

As usual, a thorough and thought-provoking reply. I feel I'm on a firmer foundation now. Thank you.

Russ

[Stefan B]

Furthermore, being able to see detailed, consistent diffraction patterns in stellar profiles, yields a lot of clues as to the veracity of detail, focus, and characteristics of the optical train.

Hi Ivo,

that's interesting. I've been struggling with this halo/fringe/diffraction thing quite a bit in my images. I don't think it's a processing artifact since I am pretty sure it's in the data. In the mean time a got used to it, but it certainly looks different compared to the majority of other images obtained with similar setups.

But what can I learn from the diffraction patterns in particular? In the image below many stars have blue "fringes". The mostly blue color of the stars is certainly caused by the usage of a duo NB filter. The fact that the center of the stars is white is due to them being overexposed? That would surprise me since many of the stars aren't very bright.

And what about this broadband image with duo NB data enhanced by NBAccent:

Can you see anything in the diffraction pattern that tells you there's something wrong with the image train (apart from the tilt that messed up the corner in the lower left)? Are there processing errors on my side which messed up stellar profiles or do they look "documentary"?

I would really appreaciate your input so I am able to better analyse my images

Regards
Stefan

[UlfG]

Hi, following with great interest. I have exactly the same imaging setup as Stefan, and very much the same issues and questions.
/Ulf

[Mike in Rancho]

Stefan, that does in fact sound similar to a prior convo? Or perhaps I'm imagining it lol.

Your stars look great to me, what do you think is wrong with them? Now that could be because the overwhelming bulk of my processing experience is with ST, so in my eye "ST stars" are normal. I have very little experience with traditional histogram levels+curves processing and blowing up stars, but I've seen it around.

Is that still the MPCC? If so you have tamed it far better than I have been able to do with mine so far. A few slightly non-circular star shapes, and on the left edge of the one image there is almost a coma-like effect in the red. The seem fine to me otherwise!

Putting a pixel value pointer on the cores of your stars, many are at triple 255's, moreso in the one image although in the other they are pretty close. So, at least as stretched for the final image, they are about as saturated as they can get. Or are already so.

Are you saying that they were not that way following AutoDev? If they got that way from using SVD, which is quite possible as it can have a brightening effect (and still sometimes that bright white ring as well), perhaps the increase dynamic range setting could help pull it back a bit?

[Stefan B]

Stefan, that does in fact sound similar to a prior convo? Or perhaps I'm imagining it lol.

Mike, we definitely discussed things like "why do my stars are so blue when doing HOO". The answer was that usually stars are much brighter in the OIII wavelength than in the Ha wavelength. Now it's rather the question "why aren't my stars blue overall but instead white in the center"

Putting a pixel value pointer on the cores of your stars, many are at triple 255's, moreso in the one image although in the other they are pretty close. So, at least as stretched for the final image, they are about as saturated as they can get. Or are already so.

Mmhhh...just tested on some of the NGC 7000 stars. Even the brighter ones aren't anywhere near 255 in the red so they still contain information. Actually they contain a surprisingly small amount of red so I would have thought the centers should appear more blue. The RGB ratios of the white centers don't seem much different from the blue "outskirts". Mmmhh...

By the way, looking at individual subs PS RAW isn't giving an overexposing warning on many of the stars which appear blue fringed in the final image. My understanding is that ST does not bloat stars if using AutoDev. So I shouldn't even be able to do it

Your stars look great to me, what do you think is wrong with them?

Look at the two blue stars in the upper center of the Heart Nebula image. The have nice spikes and a well defined center...aaaand a blue, round diffraction around the center. I don't see this much in other images. At least not to that extent. Does that indicate that something's wrong with my imaging train or is it normal signal (albeit just diffracted light) interpreted as detail by ST?

Is that still the MPCC? If so you have tamed it far better than I have been able to do with mine so far. A few slightly non-circular star shapes, and on the left edge of the one image there is almost a coma-like effect in the red. The seem fine to me otherwise!

Yes, still MPCC, never used another one. Basically, I am fine with my star shapes (and I think my f5 setup is much more foregiving than your f4 one) but this fringing thing keeps me frequently thinking if there's something wrong...

Regards
Stefan

[admin]

Furthermore, being able to see detailed, consistent diffraction patterns in stellar profiles, yields a lot of clues as to the veracity of detail, focus, and characteristics of the optical train.

what can I learn from the diffraction patterns in particular? In the image below many stars have blue "fringes". The mostly blue color of the stars is certainly caused by the usage of a duo NB filter.

With the image at hand, I cannot readily observe detailed diffraction patterns (which is a useful cue in itself!). At full resolution, there appears some blotchiness to the outlines of stars, and it is clear the stacker was unable to properly align the two channels (causing differently colored fringing).

If the alignment issues can be overcome, then we can proceed to interpret the current PSF (e.g. diffraction of any random point light in the image). Currently that diffraction pattern shows single, non-over-exposing point lights taking up many pixels. If deconvolution cannot efficiently coalesce such "blurry points" into their original point lights, then your image is needlessly oversampled.

If your image is needlessly oversampled, you are leaving signal on the table that could be put to better use; it could be used to allow decon to better resolve detail (and - in the process - diffraction patterns).

The fact that the center of the stars is white is due to them being overexposed? That would surprise me since many of the stars aren't very bright.

You don't need star cores to be overexposed to appear colorless! You just need - after color balancing of course - star cores to be of equal brightness across all channels (as an 8-bit example, R:G:B 64:64:64 will appear just as colorless as R:G:B: 255:255:255). No matter how little or much you stretch something that is colorless, it will remain colorless.

Of lesser importance, but still relevant; while StarTools tries its best to recover color information, there is also a point where color information becomes unreliable in the highlights, and it is better to "fade to white". Some cameras and/or stackers combinations introduce exacerbated aberrant color information in the highlights. This is mostly due to the RAW converter process not definining well what should happen when one channel over-exposes, but not the other(s). However, the same issue still crops up when compositing multiple mono stacks/channels in StarTools. In both cases the Color module handles this by progressively desaturating such "unreliable"/"undefined" areas/singularities.

And what about this broadband image with duo NB data enhanced by NBAccent:

Can you see anything in the diffraction pattern that tells you there's something wrong with the image train (apart from the tilt that messed up the corner in the lower left)? Are there processing errors on my side which messed up stellar profiles or do they look "documentary"?

The stars in this NB-augmented image appear quite good to me given the (apparent) dataset. In terms of being of documentary value, the star coloring is consistent with the stars' actual temperatures, they are easy to see, and I can see all temperatures represented (that is despite the NB augmentation).

It is often no accident that blue stars are prominent in many HII areas, given that they are stellar nurseries. Fueled by massive amounts of hydrogen, they are extremely large stars. And they live short, violent lives.

As said, there are some acquisition issues that probably warrant your attention.

If, for whatever reason, you don't like colored stars, you can of course always use the Artistic mode or progressively desaturate the highlights.

Does that help?

[Mike in Rancho]

Hey Stefan,

Were you looking at the finished versions on the pointer info? If I use Gimp to look at, say, the far left star in 7000, the central core is 235, 251, 251. A tiny bit out is a bit of a ring at 241, 255, 255. Then as you move outward it starts fading of course. Now, those may not be pure glaring white of 255, 255, 255, but they are still pretty darn bright! You can use the color chooser, or even open a new screen and bucket fill it with either of those custom RGB combos to see just how close to white it really is (though indeed there is just the slightest blue tint!). It could also be relativistic and perceptual. Meaning, all by itself you might be able to discern that 241, 255, 255 is just a wee bit blue-ish. But, in the middle of an image with other (and darker) stuff around it, it probably looks like plain and purse bright white.

Maybe?

I did look at your images some more though and I think I follow what you are asking. I tried to look at a few of my own, but with only moderate success. I just have very limited time with the Newt so far, and am really a rank beginner trying to figure out collimation and whatnot. Not to mention I think I misaligned my vanes and so have to go fix that. Oh and have I mentioned that this thing has more light leaks than I can count! So I am working on that too as all my images so far have had gruesome gradients.

That said, I think it's a couple things here. One is sort of the diffraction "halo" (just for Ivo!) around the star - the smaller spikes radiating out in all directions. Perhaps depending on seeing and atmosphere, that region dances around a bit, thus filling it in as sort of a ball around the star. We also know that ST resolves the stars very well via AutoDev, SVD, and so on, as much towards the pixel point source as possible. Compare that to some other techniques and software, where if you put them side by side you will see that the whole "ball" region, which in ST contains quite a bit of color, is bloated and white.

I took a stack of mine into Siril. Not that I really know what I am doing lol. But I ran the photo-color-thingy, then the gradient extraction (didn't work so well), then a asin, then the auto histogram stretch. It still looks passable and the stars are okay. But as said they are all bright white across the same expanse where ST has pinpointed the core and brought color to the diffracted area, which is much dimmer. ST's result also seems to help resolve some of the diffractive spiking, and you can also make out close doubles, which are lost in the bright bloat of the other software.

In some cases, I think it could also be the background behind a star (let's say reddish nebula) that makes that diffraction ball stand out much more.

Here's one of my recent acquisitions. Very much a work in progress and with all the errors I noted above, but as close to yours as I think I can get, Newt plus MPCC. The star Hatysa seems to have both a diffraction ball as well as a true halo, perhaps the glass of the MPCC? But by and large relatively tame against a dark sky background, and decent clarity. Compared to an excerpt of the same star with Siril.

I really don't have good data yet using the L-eNhance to see how that addition may alter what we are seeing around the stars -- other than the known coloring of course.

M42 241m ST8 4A like 2h 3C.jpg (508.59 KiB) Viewed 5678 times

M42 3pt5hr Siril Hatysa crop.jpg (46.73 KiB) Viewed 5678 times

EDIT: never mind all my theorizing. Didn't realize Ivo had answered while I was still typing!

[admin]

Also, if you have never came across it in the past, this is from the AutoDev documentation and perfectly demonstrates/encapsulates what StarTools strives to accomplish;

This is a crop of a Hubble dataset, processed by a simple AutoDev. You should be observing 0 bloat of the over-exposing stars, while faint fuzzies (e.g. background galaxies) show distinct cores and detail, and can be perfectly discerned from stars, by simply looking for the diffraction patterns of the latter.