51 replies to this topic

[EPinNC]

Overall, I'm fairly happy with this image:

 

 

Large

 

But look at some of these core stars:

 

 

I can understand bright stars being saturated and thus white, as many are.  But why these weird yellow stars (and sometimes a bit of green)?

• AT102ED refractor (doublet)
• unmodified Nikon D5600
• 30-second exposures
• 361 subs (3 hours)
• Stacked/stretched in Siril 1.2.0
• StarNet, using multiple small GHT stretches for background, and multiple small masinh stretches for stars (masinh stretches seem to reduce bloating and blue halos from this doublet refractor).

[happylimpet]

Has something gone wrong with starnet? looks like parts of the nebula have been treated as stars.

[Borodog]

Yellow == holes in the blue channel. No idea why.

[EPinNC]

Has something gone wrong with starnet? looks like parts of the nebula have been treated as stars.

I recall that StarNet did display some hot spots where bright stars were, so it wasn't 100% successful at removing them.  I'll re-run it and post what I come up with.  Thanks!

[EPinNC]

(1)  Starless image from StarNet, stretched 4 times using Siril's GHS transformation:

 

 

Note that the Trapezium is still shown, a bit fuzzy but still just white, saturated stars.  A few other stars were not completely removed.  Nothing really unexpected.  Other than that, no real weird issues.

 

(2)  Star mask from StartNet, stretched 9 times using Siril's modified asinh transformation (under the generalized hyperbolic transformation tool):

 

 

Here you can see some yellow color developing on a couple of the bright stars in the center, while others are just white/saturated.

 

When I recombine the two fully-stretched image, I get what my original image shows (aside from some minor enhancements done later.

 

All work was done at 32 bit color depth.

[Spaceman 56]

what was the verdict EPinNC ?

 

what caused these yellow stars ?

[imtl]

These are not yellow stars. These are processing artifacts from not controlling stretching very well. If combined with the use of starnet then one would get stuff like this.

[BQ Octantis]

The problem is in how you stretched your stars—the reducing of the blue bloating via your multiple asinh stretches also sacrificed some of the blue in many stars. Since what remains is green+red, they are yellow.

 

You would get better results with G-RGB stars stretched with a single asinh or GHT IMO.

 

BQ

[EPinNC]

These are not yellow stars. These are processing artifacts from not controlling stretching very well. If combined with the use of starnet then one would get stuff like this.

Definitely.  I tried a few modified asinh stretches on the un-StarNet'ted image just as a trial.  No artifacts appeared.  Something I'm doing wrong with star stretching.  Thank you!

 

 

The problem is in how you stretched your stars—the reducing of the blue bloating via your multiple asinh stretches also sacrificed some of the blue in many stars. Since what remains is green+red, they are yellow.

 

You would get better results with G-RGB stars stretched with a single asinh or GHT IMO.

 

BQ

Gotcha.  Thanks!  What are "G-RGB" stars?

 

I have tried GHT on stars alone, but they have always ended up very bloated.  It seemed that the modified asinh stretch (as part of the GHT tool) gave me tighter stars.

 

I'll try some more stuff today.  Thanks for the input, everyone!
 

[BQ Octantis]

Gotcha.  Thanks!  What are "G-RGB" stars?

 

In Siril

 

1. Image Processing → Extraction → Split Channels…
2. Image Processing → RGB Compositing… → files R, G, B
3. Pick the G channel file for L, then R for R, G for G, B for B

 

Now you have a G-RGB image without blue-bloated stars. Stretch with one asinh or GHT and see what you get.

 

BQ

[imtl]

Definitely. I tried a few modified asinh stretches on the un-StarNet'ted image just as a trial. No artifacts appeared. Something I'm doing wrong with star stretching. Thank you!


Gotcha. Thanks! What are "G-RGB" stars?

I have tried GHT on stars alone, but they have always ended up very bloated. It seemed that the modified asinh stretch (as part of the GHT tool) gave me tighter stars.

I'll try some more stuff today. Thanks for the input, everyone!

He meant that you use G for L.

[EPinNC]

In Siril

 

1. Image Processing → Extraction → Split Channels…
2. Image Processing → RGB Compositing… → files R, G, B
3. Pick the G channel file for L, then R for R, G for G, B for B

 

Now you have a G-RGB image without blue-bloated stars. Stretch with one asinh or GHT and see what you get.

 

BQ

 

Thanks so much BQ!  I'll walk through what I believe is what you are suggesting.

 

Here's the core area from the unstretched starmask produced by running StarNet Star Removal (with the "Pre-stretch linear image" option checked), with the initial histogram from GHT just for reference:

 

 

I run Image Processing > Extraction > Split Channels, and give names to each channel.  I now have the following files:

• blue.fit
• green.fit
• red.fit

Next, I run Image Processing > RGB Compositing, selecting the files accordingly:

 

 

I don't know if I need to do any colour balancing, but I'm pretty sure that I don't need to align these images as they are derived from the same original image.  So, I don't do anything here except click the "Close" button.

 

I am now looking at an "Unsaved compositing result", and I presume it is ready to stretch.

 

If I now try a single "Asinh Transformation", pushing the Stretch factor all the way to 1000.0 (just to see what's going on), I see a bunch of blue (and green) stars and other star-like artifacts:

 

 

Clearly not the result I want.  I'll try a Modified Asinh stretch and then a GHT stretch...

 

Opening the Generalised Hyperbolic Stretch Transformations dialog shows this to begin with:

 

 

The red channel is clearly reaching saturation, and green and blue are almost there as well.  It's going to be difficult to avoid at least some star saturation here.

 

Trying a "Modified arcsinh transform" (Independent channel values) to a moderate degree (Stretch factor = 8.000) clearly shows the artifacts;

 

 

Using the "Generalised hyperbolic transform", I try a relatively mild stretch (Stretch factor = 5.000, Local stretch intensity = 5.000), I see much the same thing:

 

 

I've tried a bunch of different settings here, and if I don't stretch too much the non-star artifacts are less visible, but that one star is still at least slightly yellow.  When I recombine the stars with the background image, the yellow color somehow becomes more obvious.  And the overall image just doesn't have as many stars visible as I'd like.

 

I'm a bit stumped.  Could this just be an effect of overexposure of this extremely bright core region?

 

Thanks again for any suggestions you (all) may have!

 

If anyone would like to take a shot at processing my unstretched stack:

 

https://drive.google...?usp=drive_link

 

File name:  Messier-42-30s-f7-iso200-714mm-361frames-round0.694-wFWHM-stacked.fit (about 290 MB)

 

AT102ED doublet refractor, focal length = 714mm, f/7
Nikon D5600, DSLR camera (unmodified), pixel size detected as 3.92μm

[BQ Octantis]

Wow!

 

(Click for full size)

 

There were no stars with yellow cores.

 

BQ

[EPinNC]

Wow!

 

(Click for full size)

 

 

There were no stars with yellow cores.

 

BQ

Nice!  Now, how did you stretch the stars?  Did you do anything significantly different from what I described earlier?

 

(something something give someone a fish vs. teach them to fish... )

[BQ Octantis]

If only it were fishing! Instead, I go into a dopamine-inducing, trance-like state . The better the result, the more the dopamine—so like a junky, I keep going, throwing all of my tools at it. When I hit the end, I emerge from the state…and the lights are out and everyone's gone to bed.

 

Incidentally, I taught my foreign exchange student how to fish. It was at my parents pond in the Texas Hill Country, which in 2010 was going dry (it's totally dry now). I like topwater lures, so that was my first lesson: I assigned her my Heddon Tiny Torpedo. With it, she reeled in the biggest largemouth bass I'd ever seen come out of that pond on her first cast ! But on my first cast with my Popper, I reeled two equally-sized fish, one on each treble. Turns out, fishing where the fish are starving and hitting anything that touches the water is a great way to spoil a first-time fisherman!

 

Let me see if I can derive something much simpler momentarily…

 

BQ

[BQ Octantis]

OK, I think I've got a very workable, reasonably short workflow that produces excellent color results on stars and nebula without starnet:

 

1. White Balance
2. Background subtraction
3. Histogram conditioning
4. Asinh stretch & color boost

 

 

First, we need the accurate white balance factors for the camera. Since you have a stock mainstream DSLR, this is mindlessly easy:

 

https://www.dxomark....s#measuretabs-7

 

So the white balance factors are kR_raw = 2.04 and kB_raw = 1.47.

 

With pixel math, you can apply those to the split channels. I hate using the variables R, G, B for anything but sRGB, so I use D, E, F. Siril doesn't care—it calls them Channel 0, Channel 1, Channel 2—similar to Python. So that looks like this:

 

 

This feels way more like a cooking class in a biochemistry lab…are you with me so far? If so, I'll proceed to Step 2.

[EPinNC]

Ok, before you get too far...  No, I'm not really with you.  I'm more at the stage of Chumley the Walrus in this 1-minute YouTube clip:

 

https://www.youtube....h?v=D3OSTflMO80

 

Let's back up.  When you say "without starnet", are you saying you have a method to stretch both stars and background nebulosity without actually even using StarNet?  I think that would be of considerable interest as a standalone post, if that's what you mean.

[BQ Octantis]

Ok, before you get too far...  No, I'm not really with you.  I'm more at the stage of Chumley the Walrus in this 1-minute YouTube clip:

 

https://www.youtube....h?v=D3OSTflMO80

 

Let's back up.  When you say "without starnet", are you saying you have a method to stretch both stars and background nebulosity without actually even using StarNet?  I think that would be of considerable interest as a standalone post, if that's what you mean.

 

Ok, I clicked.

 

Yes, just a unified workflow. That is the result in Post #16.

 

Let me know when you get through Step 1…

 

BQ

Edited by BQ Octantis, 04 February 2024 - 08:44 AM.

[EPinNC]

Ok, I clicked.

 

Yes, just a unified workflow. That is the result in Post #16.

 

Let me know when you get through Step 1…

 

BQ

Here's what I've done so far (documenting for both myself and others):

 

Open unstretched stack, in my case cropped, with PCC and background extraction already run.

 

Image Processing menu > Extraction > Split Channels

 

Color Space:  RGB

Red:  R
Green:  G
Blue:  B

 

Click "Apply".  This produces 3 files:  R.fit, G.fit, and B.fit

 

Click "Close".

 

Image Processing menu > Pixel Math

 

(see PixelMath documentation for how to navigate this dialog)

 

Uncheck "Use single RGB/K expression"

 

In each of the first three boxes labeled "R", "G", and B", enter a variable name, as simple as "R", "G", and "B".

 

Click the "+" button at the upper right of the dialog box, and add R.fit file.  Do the same for G.fit and B.fit.  The full paths to the selected files will appear in the "Images" pane.  Note the "Variable" names assigned to each .fit file.

 

In the "Parameters" box, type the variable assignment equations specified on the page for the D5600 that BQ linked to for "White balance scales".

 

Click "Apply".  Now see the image turn a spectacular magenta color:

 

What's next, Professor Whoopee?

[BQ Octantis]

The inputs you gave Siril are producing a constant vice a multiplier on the pixel values.

 

Instead, change the values in the Parameters field to

 

kR = 2.04*R, kB = 1.47*B

 

Then change the R field to

 

kR

 

and the B field to

 

kB

 

That will get you to the end of Step 1, and you should be able to make out the nebula (in a sea of red light pollution).

Edited by BQ Octantis, 04 February 2024 - 01:09 PM.

[EPinNC]

Doh.  I didn't look closely enough at your example.  Thank you!

 

[BQ Octantis]

The math is right, but I'm worried the image doesn't look like mine. What screen preview mode are you using? I specifically used Autostretch. Are you starting with the unprocessed raw channels, i.e., extracted from your Google Drive fit file?

Edited by BQ Octantis, 04 February 2024 - 03:18 PM.

[EPinNC]

The math is right, but I'm worried the image doesn't look like mine. What screen preview mode are you using? I specifically used Autostretch. Are you starting with the unprocessed raw channels, i.e., extracted from your Google Drive fit file?

That was an autostretch (preview).  I started with an image that had already been cropped, photometric color calibration, and background extraction.  I tried it again with the raw, unprocessed stack:

 

 

 

Now that it's taken me 3 tries to reproduce what you did, are you ready for Step 2? 

Edited by EPinNC, 04 February 2024 - 03:44 PM.

[BQ Octantis]

YAY! Yes, of course!

 

Steps 1 and 2 are specifically to deal with color balance out of a DSLR, which is different to an astrocam. It replaces Photometric Color Calibration (PCC).

 

Step 1 sets the white balance factors to calibrated D50, which are reasonably good for AP.

 

For Step 2, we're subtracting the background (vice the background extraction performed by PCC in Siril). But we need a background reference, which in such a closeup of M42 is a bit challenging.

 

Switch the preview mode to Histogram and look for the dimmest part of the image. I place it in the upper right corner:

 

 

If you zoom way in, you can see there are several spots in which you can put a box to make a measurement. Put a 50×50 box among the stars, right click, and pick Statistics… and you'll see the background values:

 

 

We're after the mean, which are r=179.9, g=87.6, b=129.5. And the median values are the same, which means the spot is pretty uniform. No spot in the sky is totally black (and image processing chokes on anything zero or below), so we'll need to leave a small sky bias, which I'll call i. I like to go with i=10.

 

To subtract the background, we just go back to our Pixel Math window and put those in as parameters. But our r,g,b pixel values are on the interval [0,65535], while Pixel Math uses inputs on the interval [0,1]. So we have to divide the values we want to subtract by M=65535.

 

So the parameters field will now be

 

kR = 2.04*R, kB=1.47*B, r=179.9, g=87.6, b=129.5, i=10, M=65535

 

And our R, G, B fields will be

 

R - (r-i)/M

G - (g-i)/M

B - (b-i)/M

 

Once you apply the Pixel Math, you'll see the Statistics… for the square go to r=10, g=10, b=10. And you'll find the Autostretch image to be most satisfactory:

 

 

With me so far?

Edited by BQ Octantis, 04 February 2024 - 05:36 PM.

[Spaceman 56]

I run Image Processing > Extraction > Split Channels, and give names to each channel.  I now have the following files:

• blue.fit
• green.fit
• red.fit

Next, I run Image Processing > RGB Compositing, selecting the files accordingly:

 

 

then you select green as a luminance, and put it all back together.  I have never done this. 

 

does it work ? 

 

what is achieved ?

 

experts please advise.