Star Colour Fringing Problem

[MikeOates]

I am having an odd problem, it's not new, I just want to get to the bottom of it so that my image quality can be improved.

The red filtered images do not register with the green and blue, even though I use just one master frame (in this case a blue image) to register all the different filtered image to.

The result when combined to an RGB image is a colour fringe of red at one side of a star and blue at the other. At first you would think it an optical problem resulting from coma error, but it's not. All stars in the image have the colour fringe at the same side, the fringe does not vary in the corners as you would expect for coma.

Focusing is also well controlled and I focus through the filter being used, initial focus with a Bahtinov mask then the rest of the session using the autofocus in SGP. Filters are Astrodon.

If I manually blink the three images the green and blue images are lined up, but the red shifts one or two pixels. What is more odd is the brighter stars seen to move a larger offset than fainter ones.

Has anyone come across this issue? are settings that need changing in StarAlignment? Or is that an optical problem?

The scope is a Takahashi FSQ-106ED & mono CCD camera.

The way I have 'corrected this' before is to reduce star sizes with a mask and MT, but I really want to sort this out properly.

The rgb image here is a crop made just after combine, I have made it none linear and enhanced the colour saturation to make it more obvious. The three other small crops are one from each filter after StarAlignment.

Thanks,

Mike

[oldwexi]

Hi Mike!
Tried to find a solution.
Have a look here:
http://www.werbeagentur.org/oldwexi/StarAlignment.jpg
Registration model change to 2-D Suface Splines does step One.
Step 2 ChannelMatch moves the complete red channel  1/2 Pixel downward.

 
Anyhow, i have no idea why the small stars in your Image02 are correct(!!) in place
but the very bright ones are far off. Maybe filter quality? Dont know.

Gerald

[MikeOates]

Gerald,

Oh thank you so much for taking the time on this problem, that is a huge improvement! far better than I got before with MT.

It is certainly an odd problem.

Mike

[jkmorse]

Mike,

For what its worth, I had this issue recently when I got my CCD back from repairs.  Turns out the repair shop put my filters in wrong so I was actually putting the wrong images into the various slots.  Does not seem to be happening here because other than the stars the other colors look correct, but maybe worth checking the basics.  Also, it wasn't clear from your note whether you are refocusing between each color.  While the Astrodon filters are great and are 1:1:1 parfocal, that really only works with mirrors.  When going through lenses, like my NP101is or your 106ED, there are inevitably offsets you need to account for.  But you probably already knew that.

As I said, for what it's worth.

Jim

[MikeOates]

Hi Jim,

Yes where I say "I focus through the filter being used" I mean I focus using the same filter that I image with. But I do have filter offsets set up SGP so I do know there are a few steps different with each filter.

Mike

[MikeOates]

I have found out where the problem lies with the apparent movement of the larger stars compared to the small ones.

All the separate subs that show the movement after StarAlignment have been taken on the opposite side of the meridian compared to the sub  used as the register.

I tried doing manual 180 deg rotations of those subs with both DynamicCrop and FastRotation before StarAlignment, but this does not make any difference.

The reason the small stars appear to stay still after integration, I believe is down to the rejection working better on the smaller less saturated stars. If you blink the individual subs after StarAlignment you will see the larger stars move, and the fainter small stars are made into a short line, i.e. the stars are elongated. This is the same even when large and small stars are close together.

The result are image of stars that are in different positions, it could be that all the red subs were taken on one side and the blue on the other side of the meridian, giving this colour fringing effect. The shifting of one colour with ChannelMatch is just masking the issue and is not really a solution.

I have tried other settings in StarAlignment and this can be made much worse using 2-D Surface Splines. But I do not have the knowlege to try different settings to cure this issue, which can't be an optical problem. The camera is not rotated when passing over the meridian. And the issue is with any of my images, not just the original image I posted about.

The galaxy group subs below are of the Coma Galaxy Cluster ACO 1656, taken in NIR and are 1800s exposures. There are an awful lot of galaxies in this image.

These three subs are all calibrated, the first two were taken with the scope on one side of the meridian and the last one on the opposite side:

http://www.mikeoates.org/astro/aco_1656_1800s_nir_bin1-25c_20150324_04_c_c_cc.xisf
http://www.mikeoates.org/astro/aco_1656_1800s_nir_bin1-25c_20150324_05_c_c_cc.xisf
http://www.mikeoates.org/astro/aco_1656_1800s_nir_bin1-25c_20150324_06_c_c_cc.xisf

Use the first one as the register and you will see using Blink, the issue I have.

If this can be resolved, the quality of my images will increase a great deal, not only for the colour fringing but also the definition and increased SNR as lots of good photons are being thrown away.

I hope Juan sees this and can help.

Thank you,

Mike

[gvanhau]

Hello
I've looked at the images in the first post, and I can see there is as you also say, an elogantion of all stars in the same angle of the color fringing.
In the color image, if you take a close look, yo can also see fringing in the smaller stars, It is less visible because it is also dimmer.
I think there is little that Juan can do about it.
In my opinion, the only rigth thing to do is to enhance the capture.

Of course, there are some tricks that can be done. For instance, you could try motion blur in deconvolution, but at the end, the best is to solve the problems at the root.

Geert

[MikeOates]

Hi Geert,

I think the problem lies with StarAlignment, or the settings used. The images posted here have used StarAlignment with the default settings. The stars are pretty round, the only difference is some images were taken on one side of the meridian, and others on the other side. Even images that were taken on the same night within minutes using the same focus, filter and observing conditions.

I have been experimenting with lots of different settings in StarAlignment with no improvement. If I use StarAlignment on all the images taken on one side they all match up if the reference image used is also on the same side. I can then do the same for all the other images, again with the reference image from the same side and they all line up, but I can't line up both sets, it just does not make any sense. I have even integrated each set separately, then used StarAlignment to align both images, still the same result. Now this is really odd as by the time each set is integrated each has nice round stars. I can also see just at looking at each integrated set that they are sharper than when I had previously integrated all together in one go.

One set has 24 subs, the other 20 subs.

Here are three integrations:
Set1 (24 subs) http://www.mikeoates.org/astro/integration_set1_24.xisf
Set1 (20 subs) http://www.mikeoates.org/astro/integration_set2_20.xisf
All 44 subs http://www.mikeoates.org/astro/integration_all_44.xisf

Even if I could get the above two sets aligned, that is the wrong way to do this, as I am loosing out on the rejection advantage of lots of subs. There are satellite and airplane trails left in the images which are not present in the full set of 44 subs integrated in one go.

Getting this right is crucial if I am to get the best results, thank you,

Mike

[Juan Conejero]

Hi Mike,

The problem is in the "..._20150324_06" image, that is, the one that is rotated by 180 degrees with respect to _04 and _05. This image has a clear tracking issue. If you have images like this one in your data set, then this explains the color fringing problem that you have. There is not much we can do (without entering the cheating field) to fix this with image processing. Apparently, something bad happens to your tracking system when you perform a meridian flip. Could be a mount balancing problem, flexure, poor fastenings, sliding optics, deformed optics, etc.

[MikeOates]

Hi Juan,

Thank you very much for looking into this for me, it's been driving me mad.

Perhaps I showed the wrong image as an example, most of the other subs are fine and not trailed. I just wanted to place two images taken one after the other to show it can't be the observing conditions.

What your saying still does not make sense to me if you looked at the second examples where I was trying to align two integrated images. Those, due to the integration, now have round stars but won't align.

But moving on, I have another example of a pair of subs to show you, both on the observing session, just not next to each other in time. Both of these have no tracking issues and have very good round stars (600s exposure). Again each were taken on different sides of the meridan. These also show the larger stars moving after using StarAlignment and viewed with Blink.

I do hope you can download these, do the alignment and check the result. It does not matter which image you use as the reference, the result is the same.

http://www.mikeoates.org/astro/coddingtons_nebula_blue_bin1-25c_600s_20160114_11_c_cc.xisf

http://www.mikeoates.org/astro/coddingtons_nebula_blue_bin1-25c_600s_20160114_13_c_cc.xisf

Thank you,

Mike

[Juan Conejero]

Hi Mike,

I have downloaded your images, and have tested StarAlignment thoroughly with them. SA registers the images correctly. I have verified this by measuring centroid positions with DynamicPSF, on the reference and registered images (11 and 13 in my test, respectively). The differences are well within the expected tolerances for most stars (about 0.1 - 0.2 px, which is in good agreement with the reported RMS error of 0.178 px). However, for the brightest stars there is a systematic difference of about one pixel on the vertical axis, fitting elliptical PSF model functions. This causes the fringing problem.

Another test to verify registration accuracy is running StarAlignment in matched stars mode on full previews defined on the reference and target images. I have done this, and have been unable to find a single case where the matched centroid is not located at its optimal position.

In my opinion, there is something wrong with acquisition in these images, and the problem is obviously related to meridian flipping. The issue is very noticeable on bright stars, but barely visible on the dimmest ones. StarAlignment finds the best compromise solution by minimizing differences for most stars, and does the job well, as expected. This is a very interesting problem from an image processing perspective, but of course I understand your exasperation. My advice is to concentrate in finding a solution in the hardware/acquisition front, and use image processing as an analysis resource, just as we have done in this example.

Let's see if other users with more practical experience (remember that I am a sort of laboratory freak) can add their points of view.

[MikeOates]

Juan,

Thank you very much for looking further into my problem, it is most appreciated.

I will look in the hardware side of things, in fact I already have plans to replace the guide scope and reduce the number of adapter / spacing rings in the imaging scope optical path to make it more rigid. But as Juan says if anyone else can offer any advice, especially if you have had a similar problem that would be great.

Mike

[NKV]

Hi Mike,

Ideally, all your filters must be parallel to CCD, but in real world, carousel filter wheel can set a filters in a bit different angles.
Especially after meridian flip.   
So, the problem in reflection from optics. I think reflection from filters and it's normal to see reflection only from very bright stars.

[MikeOates]

Thank you, that's an interesting idea, not sure what I can do about at the moment. But making the above planned changes to the spacers I have will change the position of the filter wheel in relation to the camera, i.e. closer, so that may have an effect.

Mike