Problem with StarXTerminator in dense star fields

0.90 Astrogerdt ... · view views · Share link
Hello guys, I have some problems with star removal using either StarXTerminator or StarNet V2 for some time now. Maybe someone on here has a solution for this. My workflow is as follows: - Integrate using WBPP - GradientCorrection - SPCC - BlurXTerminator - Star removal (either SXT or StarNet V2) Removing the Stars after I applied BXT leaves more artifacts in the center, but fewer artifacts near the edges. Earlier I did wide field images with a dual setup with some 50mm lenses that were rather low quality. The star removal with both programs always left some ugly artifacts, but I blamed the low quality lenses and bad star shapes for this. I hoped getting better lenses would solve this problem in the future. Now I upgraded to 2x Samyang 135mm F2 with relatively good quality. Unfortunately, I still run into the same issue with star removal, this time probably because of the insanely high star density in the frames instead of low star quality. Here is an example: Image with stars: Workflow: WBPP, GradientCorrection, SPCC, BXT Image without stars: Workflow: Applied SXT with "large overlap" option to the prior image. Both images have the same STF applied to them and are resampled and compressed to make the file size fit for the forum. Here is a 200% comparison of a representative star field: As you can see, the starless image shows a big dark region and has lots of point like artifacts in it. Regarding the big dark region, I can't be absolutely sure that it hasn't been there prior to BXT, but I can't see it even when I stretch the image with stars very extremely. So I am pretty certain that this artifact is introduced by SXT. As for the small point like artifacts, I identify them as leftovers from very small stars. StarNet V2 does an even worse job with this. In general, I assume that all these artifacts arise due to the crazy star density in the image. It is probably close to impossible to get a good background estimate in these regions due to this. But it still limits the processing possibilities for such fields. Does anyone of you know a fix for this problem, or has any comments on this? I already wrote with Rus about this, and he said that future versions will address this fix, but he does not know how long it takes to develop them, and it has already taken a decent time. So I would like to have a solution until the development is done. CS Gerrit Like

0.90

...

Hello guys,

I have some problems with star removal using either StarXTerminator or StarNet V2 for some time now. Maybe someone on here has a solution for this.

My workflow is as follows:
- Integrate using WBPP
- GradientCorrection
- SPCC
- BlurXTerminator
- Star removal (either SXT or StarNet V2)

Removing the Stars after I applied BXT leaves more artifacts in the center, but fewer artifacts near the edges.

Earlier I did wide field images with a dual setup with some 50mm lenses that were rather low quality. The star removal with both programs always left some ugly artifacts, but I blamed the low quality lenses and bad star shapes for this. I hoped getting better lenses would solve this problem in the future.

Now I upgraded to 2x Samyang 135mm F2 with relatively good quality. Unfortunately, I still run into the same issue with star removal, this time probably because of the insanely high star density in the frames instead of low star quality. Here is an example:

Image with stars:

Workflow: WBPP, GradientCorrection, SPCC, BXT

Image without stars:

Workflow: Applied SXT with "large overlap" option to the prior image.

Both images have the same STF applied to them and are resampled and compressed to make the file size fit for the forum.

Here is a 200% comparison of a representative star field:

As you can see, the starless image shows a big dark region and has lots of point like artifacts in it.
Regarding the big dark region, I can't be absolutely sure that it hasn't been there prior to BXT, but I can't see it even when I stretch the image with stars very extremely. So I am pretty certain that this artifact is introduced by SXT.
As for the small point like artifacts, I identify them as leftovers from very small stars.
StarNet V2 does an even worse job with this.

In general, I assume that all these artifacts arise due to the crazy star density in the image. It is probably close to impossible to get a good background estimate in these regions due to this. But it still limits the processing possibilities for such fields.

Does anyone of you know a fix for this problem, or has any comments on this?

I already wrote with Rus about this, and he said that future versions will address this fix, but he does not know how long it takes to develop them, and it has already taken a decent time. So I would like to have a solution until the development is done.

CS Gerrit

9.89 andrea tasselli ... · view views · 3 likes · Share link
First off, I can't see anything wrong with the starless image unless you plan to use just that, which obviously would be foolish. If you were to stretch both starless and stars separately, which I assume you do want to, then the "holes" will disapper once the stars are back in place. Secondly, if you were to operate BXT with a stretched version of the image then BXT would do a better job at leaving the starless relatively unscathed. Besides, I processed far denser fields than this with no major drawback from my point of view. And, yes, the big dark(ish) region was there all the time. Like

9.89

andrea tasselli

...

· 3 likes

· Share link

First off, I can't see anything wrong with the starless image unless you plan to use just that, which obviously would be foolish. If you were to stretch both starless and stars separately, which I assume you do want to, then the "holes" will disapper once the stars are back in place. Secondly, if you were to operate BXT with a stretched version of the image then BXT would do a better job at leaving the starless relatively unscathed. Besides, I processed far denser fields than this with no major drawback from my point of view. And, yes, the big dark(ish) region was there all the time.

7.19 Dan Kearl ... · view views · 2 likes · Share link
I think SXT did an amazing job on this, I have no idea what you think is wrong. Like

0.90 Astrogerdt Topic starter ... · view views · Share link
Thanks for your replies! First, there is the large dark region after applying SXT. This is a large scale feature, that won't be correctly restored after bringing in the stars. As I said, I can't guarantee it, but it is most certainly not there before SXT. And I see that in many of my images. Secondly, there are some features which I visually identify as very small stars that aren't removed. This gives the background after removing the stars such a "noisy" look, outside of the larger, removed stars. This brings two problems: first, it lowers the authenticity of the data, because a lot of small stars will be way too bright after stretching the object, and secondly, any noise reduction on the starless image will try to remove these stars, as they are incorrectly treated as noise. Admittedly, for the second problem, one has to zoom in pretty much to see it, but since this is present in the complete image, I still think it is important. Unfortunately, I can't create a blinking animation of this that is uncompressed, I can only provide uncompressed images of the same region. Maybe it is better to see once you load them into a blinking tool and compare the very small stars. Here are the images: Normal: drizzle_integration_GC_SPCC_BXT_Preview05.tif Starless: drizzle_integration_GC_SPCC_BXT_SXT_Preview05.tif CS Gerrit Like

0.90

Astrogerdt

Topic starter

...

· Share link

Thanks for your replies!

First, there is the large dark region after applying SXT. This is a large scale feature, that won't be correctly restored after bringing in the stars.
As I said, I can't guarantee it, but it is most certainly not there before SXT. And I see that in many of my images.

Secondly, there are some features which I visually identify as very small stars that aren't removed. This gives the background after removing the stars such a "noisy" look, outside of the larger, removed stars.
This brings two problems: first, it lowers the authenticity of the data, because a lot of small stars will be way too bright after stretching the object, and secondly, any noise reduction on the starless image will try to remove these stars, as they are incorrectly treated as noise.

Admittedly, for the second problem, one has to zoom in pretty much to see it, but since this is present in the complete image, I still think it is important.
Unfortunately, I can't create a blinking animation of this that is uncompressed, I can only provide uncompressed images of the same region.
Maybe it is better to see once you load them into a blinking tool and compare the very small stars.

Here are the images:
Normal: drizzle_integration_GC_SPCC_BXT_Preview05.tif
Starless: drizzle_integration_GC_SPCC_BXT_SXT_Preview05.tif

CS Gerrit

9.89 andrea tasselli ... · view views · Share link
Not sure what are you complaining about. The examples you shared are rather stamp-sized and it would be best to share the original at full bit depth, but here is what it came out using SN++ V2. SXT doesn't work well unless it's full 32bit, apparently: The one one the right is your starless, mine is in the middle. Stars to the left. Like

12.96 Kay Ogetay ... · view views · Share link
Thanks for your replies! First, there is the large dark region after applying SXT. This is a large scale feature, that won't be correctly restored after bringing in the stars. As I said, I can't guarantee it, but it is most certainly not there before SXT. And I see that in many of my images. Secondly, there are some features which I visually identify as very small stars that aren't removed. This gives the background after removing the stars such a "noisy" look, outside of the larger, removed stars. This brings two problems: first, it lowers the authenticity of the data, because a lot of small stars will be way too bright after stretching the object, and secondly, any noise reduction on the starless image will try to remove these stars, as they are incorrectly treated as noise. Admittedly, for the second problem, one has to zoom in pretty much to see it, but since this is present in the complete image, I still think it is important. Unfortunately, I can't create a blinking animation of this that is uncompressed, I can only provide uncompressed images of the same region. Maybe it is better to see once you load them into a blinking tool and compare the very small stars. Here are the images: Normal: drizzle_integration_GC_SPCC_BXT_Preview05.tif Starless: drizzle_integration_GC_SPCC_BXT_SXT_Preview05.tif CS Gerrit Is it possible that the starry field is confusing? Sometimes it causes us to mispredict the background. I, myself, sometimes get surprised when I see the background image after star removal, then I realize it is correct. Not sure what you are exactly referring to as 'large dark region', but if I understand correctly, it is already there. Sometimes it is hard to see for some people, especially if you perceive colors differently. Maybe switching to grayscale might help. But I don't see any large dark regions that did not exist in the first place. Edited ... Like

12.96

Kay Ogetay

...

· Share link

Thanks for your replies!

First, there is the large dark region after applying SXT. This is a large scale feature, that won't be correctly restored after bringing in the stars.
As I said, I can't guarantee it, but it is most certainly not there before SXT. And I see that in many of my images.

Secondly, there are some features which I visually identify as very small stars that aren't removed. This gives the background after removing the stars such a "noisy" look, outside of the larger, removed stars.
This brings two problems: first, it lowers the authenticity of the data, because a lot of small stars will be way too bright after stretching the object, and secondly, any noise reduction on the starless image will try to remove these stars, as they are incorrectly treated as noise.

Admittedly, for the second problem, one has to zoom in pretty much to see it, but since this is present in the complete image, I still think it is important.
Unfortunately, I can't create a blinking animation of this that is uncompressed, I can only provide uncompressed images of the same region.
Maybe it is better to see once you load them into a blinking tool and compare the very small stars.

Here are the images:
Normal: drizzle_integration_GC_SPCC_BXT_Preview05.tif
Starless: drizzle_integration_GC_SPCC_BXT_SXT_Preview05.tif

CS Gerrit

Is it possible that the starry field is confusing? Sometimes it causes us to mispredict the background. I, myself, sometimes get surprised when I see the background image after star removal, then I realize it is correct. Not sure what you are exactly referring to as 'large dark region', but if I understand correctly, it is already there. Sometimes it is hard to see for some people, especially if you perceive colors differently. Maybe switching to grayscale might help. But I don't see any large dark regions that did not exist in the first place.

Edited ...

0.90 Astrogerdt Topic starter ... · view views · Share link
Ok, I kind of fixed the problem in the mean time, at least as far as I can tell. I missed one of the most basic things about tiny stars: drizzle 2x or 3x! My initial image was processed with CFA Drizzle 1x, redoing all the preprocessing with Drizzle 2x solved the problem nearly perfectly. My theory is that the stars were too small and pixelated to be reliably identified as stars, so drizzling with an increased resolution solved this issue as it gave all stars a nice shape. The workflow with the best results is thus: - WBPP with Drizzle 2x - BlurXTerminator - StarXTerminator Here is a comparison between the two workflows with the only difference being the Drizzle scale: This is a 200% zoom of the preview I used. On the left side, you can see the original drizzle 2x image. In the middle is the drizzle 1x image with GC, BXT and SXT applied. On the right is the drizzle 2x image with GC, BXT and SXT applied. At least for me, the right image looks much better in terms of star removal. A lot of the faint and small stars that were not removed with the drizzle 1x image are perfectly removed now. This makes processing much easier, but also takes much more time. As for the background issue I mentioned, I am talking about these regions: These regions are, as far as I can tell, darker than they are in the original image prior to star removal. Also, even in the small preview region I used for the Drizzle 1x vs Drizzle 2x comparison, I can see a clear change in background color and brightness between both versions: This indicates to me, that even slight changes in the image parameters drastically change the background estimation of SXT. I will have to re-check this with StarNet V2. CS Gerrit Like

0.90

Astrogerdt

Topic starter

...

· Share link

Ok, I kind of fixed the problem in the mean time, at least as far as I can tell.

I missed one of the most basic things about tiny stars: drizzle 2x or 3x!

My initial image was processed with CFA Drizzle 1x, redoing all the preprocessing with Drizzle 2x solved the problem nearly perfectly.

My theory is that the stars were too small and pixelated to be reliably identified as stars, so drizzling with an increased resolution solved this issue as it gave all stars a nice shape.

The workflow with the best results is thus:
- WBPP with Drizzle 2x
- BlurXTerminator
- StarXTerminator

Here is a comparison between the two workflows with the only difference being the Drizzle scale:

This is a 200% zoom of the preview I used.
On the left side, you can see the original drizzle 2x image.
In the middle is the drizzle 1x image with GC, BXT and SXT applied.
On the right is the drizzle 2x image with GC, BXT and SXT applied.

At least for me, the right image looks much better in terms of star removal. A lot of the faint and small stars that were not removed with the drizzle 1x image are perfectly removed now. This makes processing much easier, but also takes much more time.

As for the background issue I mentioned, I am talking about these regions:

These regions are, as far as I can tell, darker than they are in the original image prior to star removal.
Also, even in the small preview region I used for the Drizzle 1x vs Drizzle 2x comparison, I can see a clear change in background color and brightness between both versions:

This indicates to me, that even slight changes in the image parameters drastically change the background estimation of SXT.

I will have to re-check this with StarNet V2.

CS Gerrit

7.22 Franco Grimoldi ... · view views · Share link
I agree with those who say that they don't see anything wrong with any of the starless images... What I do see in the first image ("Image with stars") is a noticeable gradient that doesn't seem well corrected: I set the vibrance and saturation to 100% in Photoshop to make it more evident: I see an green-bluish "cloud" going from the center to the top right of the field. Before judging the StarX or StarNet performance, it would be good to see how does the final image look like, once the stars are brought back to the scene. Like

7.22

Franco Grimoldi

...

· Share link

I agree with those who say that they don't see anything wrong with any of the starless images...

What I do see in the first image ("Image with stars") is a noticeable gradient that doesn't seem well corrected:

I set the vibrance and saturation to 100% in Photoshop to make it more evident: I see an green-bluish "cloud" going from the center to the top right of the field.

Before judging the StarX or StarNet performance, it would be good to see how does the final image look like, once the stars are brought back to the scene.

0.90 Astrogerdt Topic starter ... · view views · Share link
Ok, here is the comparison with StarNet v2: Top image is Drizzle 2x, GradientCorrection, BXT, SXT. Lower image is Drizzle 2x, GradienctCorrection, BXT, StarNet V2. In my opinion, SXT has a much better brighness preservation, but leaves a more stained background. Ideally, in my opinion again, would be a combination of the background preservation from SXT with the star removal capability of SXT. Like

0.90

Astrogerdt

Topic starter

...

· Share link

Ok, here is the comparison with StarNet v2:

Top image is Drizzle 2x, GradientCorrection, BXT, SXT.
Lower image is Drizzle 2x, GradienctCorrection, BXT, StarNet V2.

In my opinion, SXT has a much better brighness preservation, but leaves a more stained background.

Ideally, in my opinion again, would be a combination of the background preservation from SXT with the star removal capability of SXT.

0.90 Astrogerdt Topic starter ... · view views · Share link
Franco Grimoldi: I agree with those who say that they don't see anything wrong with any of the starless images... What I do see in the first image ("Image with stars") is a noticeable gradient that doesn't seem well corrected: I set the vibrance and saturation to 100% in Photoshop to make it more evident: I see an green-bluish "cloud" going from the center to the top right of the field. Before judging the StarX or StarNet performance, it would be good to see how does the final image look like, once the stars are brought back to the scene. Allright, this could be indeed a problem. Sorry, I did not see your post before posting mine. Maybe the gradient in your image could be due to using the JPG, but this seems unlikely to me. To clear things up, here are the images I was talking about: https://drive.google.com/drive/folders/1lrJi0P09PkaKTK9TTmLRdEcN-fxLYkA7?usp=sharing I am currently uploading all relevant files, it may take some time to finish. CS Gerrit Like

0.90

Astrogerdt

Topic starter

...

· Share link

Franco Grimoldi:
I agree with those who say that they don't see anything wrong with any of the starless images...

What I do see in the first image ("Image with stars") is a noticeable gradient that doesn't seem well corrected:

I set the vibrance and saturation to 100% in Photoshop to make it more evident: I see an green-bluish "cloud" going from the center to the top right of the field.

Before judging the StarX or StarNet performance, it would be good to see how does the final image look like, once the stars are brought back to the scene.

Allright, this could be indeed a problem. Sorry, I did not see your post before posting mine.

Maybe the gradient in your image could be due to using the JPG, but this seems unlikely to me.

To clear things up, here are the images I was talking about: https://drive.google.com/drive/folders/1lrJi0P09PkaKTK9TTmLRdEcN-fxLYkA7?usp=sharing

I am currently uploading all relevant files, it may take some time to finish.

CS Gerrit

9.89 andrea tasselli ... · view views · Share link
Ok, here is the comparison with StarNet v2: Top image is Drizzle 2x, GradientCorrection, BXT, SXT. Lower image is Drizzle 2x, GradienctCorrection, BXT, StarNet V2. In my opinion, SXT has a much better brighness preservation, but leaves a more stained background. Ideally, in my opinion again, would be a combination of the background preservation from SXT with the star removal capability of SXT. SXT applies a low-pass filter to the background (in a way similar to NXT but at a lower image scale I gather). Like

9.89

andrea tasselli

...

· Share link

Ok, here is the comparison with StarNet v2:

Top image is Drizzle 2x, GradientCorrection, BXT, SXT.
Lower image is Drizzle 2x, GradienctCorrection, BXT, StarNet V2.

In my opinion, SXT has a much better brighness preservation, but leaves a more stained background.

Ideally, in my opinion again, would be a combination of the background preservation from SXT with the star removal capability of SXT.

SXT applies a low-pass filter to the background (in a way similar to NXT but at a lower image scale I gather).

9.89 andrea tasselli ... · view views · Share link
This with SXT (and Graxpert before), starting from the final drizzle integration with no other processing applied: Note the squarish blocks spread all over the place and the green cast at the bottom and oversmoothed appearance. Now with SN++ V2 (still with Graxpert to remove gradient): This shows no block pattern and the distribution of color in the BG much more uniform and overall a LOT more realistic than the SXT results. Now a pass with NXT and SCNR to remove the green cast: Like

9.89

andrea tasselli

...

· Share link

This with SXT (and Graxpert before), starting from the final drizzle integration with no other processing applied:

Note the squarish blocks spread all over the place and the green cast at the bottom and oversmoothed appearance.

Now with SN++ V2 (still with Graxpert to remove gradient):

This shows no block pattern and the distribution of color in the BG much more uniform and overall a LOT more realistic than the SXT results.
Now a pass with NXT and SCNR to remove the green cast:

0.90 Astrogerdt Topic starter ... · view views · Share link
Ok, I have to admit, in the larger scale, the blotchiness I initialized took as an error looks much more like actual nebula structure in the image. Thanks guys! CS Gerrit Like

9.89 andrea tasselli ... · view views · 1 like · Share link
Just for the heck of it this the final stretched image: Like

4.51 Jan Erik Vallestad ... · view views · Share link
I'm not sure if it's been mentioned, but I don't see which camera you are using as to determine the scale of undersampling. If you didn't already drizzle more than x1 I would advice you to try x2, then resample after extracting the stars, I find that the upscaling usually helps a lot with star removal - not only when I use this lens but pretty much in general for preserving smaller objects. Also better SNR would probably help. At least that's what I've experienced myself using this lens. Like

4.51

Jan Erik Vallestad

...

· Share link

I'm not sure if it's been mentioned, but I don't see which camera you are using as to determine the scale of undersampling. If you didn't already drizzle more than x1 I would advice you to try x2, then resample after extracting the stars, I find that the upscaling usually helps a lot with star removal - not only when I use this lens but pretty much in general for preserving smaller objects. Also better SNR would probably help. At least that's what I've experienced myself using this lens.

	Distinct awarded users	Total awarded images
Image of the day
Top picks
Top pick nominations

	Image of the day	Top pick	Top pick nominations	Total submitted
Deep sky
Solar system
Extremely wide field
Star trails
Northern lights
Noctilucent clouds
Landscape
Artificial satellite

	Image of the day	Top pick	Top pick nominations	Total submitted
Backyard
Traveller
Own remote observatory
Amateur hosting facility
Public amaeteur data
Professional, scientific grade data
Mix of multiple sources
Other
Unknown

AstroBin

AstroBin

Image Index:

Contribution Index (beta):

Problem with StarXTerminator in dense star fields Russell Croman Astrophotography StarXTerminator · Astrogerdt · ... · 14 · 566 · 15

		...
You have no new notifications.

AstroBin

AstroBin

Image Index:

Contribution Index (beta):

Problem with StarXTerminator in dense star fields Russell Croman Astrophotography StarXTerminator · Astrogerdt · ... · 14 · 566 · 15

Sorry

Sorry

Sorry :-(

Sorry

Share this forum post

Post editing not available

Sorry

Sorry

Sorry :-(

Sorry

Share this forum post

Post editing not available

Sorry

Sorry

Sorry :-(

Sorry

Share this forum post

Post editing not available

Sorry

Sorry

Sorry :-(

Sorry

Share this forum post

Post editing not available

Sorry

Sorry

Sorry :-(

Sorry

Share this forum post

Post editing not available

Sorry

Sorry

Sorry :-(

Sorry

Share this forum post

Post editing not available

Sorry

Sorry

Sorry :-(

Sorry

Share this forum post

Post editing not available

Sorry

Sorry

Sorry :-(

Sorry

Share this forum post

Post editing not available

Sorry

Sorry

Sorry :-(

Sorry

Share this forum post

Post editing not available

Sorry

Sorry

Sorry :-(

Sorry

Share this forum post

Post editing not available

Sorry

Sorry

Sorry :-(

Sorry

Share this forum post

Post editing not available

Sorry

Sorry

Sorry :-(

Sorry

Share this forum post

Post editing not available

Sorry

Sorry

Sorry :-(