RGB alignment

[Ken82]

I'm hoping someone could help me out with this issue-

Basically after RGB combination using channel combination i always get strangely coloured stars. What should be a red star has a purple edge (this is always to the right hand side). I know it could possibly be CA but this is an fsq 106 so i'd expect it to be free of this issue and that its likely something i'm doing at this stage.

My process after calibration is star alignment, dynamic crop, DBE, Linear fit.

Ian King has said it may be the alignment but at this moment I've been unable to change anything by the alignment process.

Red     https://drive.google.com/file/d/18JVTbaZ62XsAlWUSlSIhN4JqtCFoSqDU/view?usp=sharing

Green     https://drive.google.com/file/d/1B2S6XDWrf2dcuWDmZLipZPp4DncRzz9S/view?usp=sharing

Blue       https://drive.google.com/file/d/1ss53Bm-oo1kn0b1XCUITeyH795HthZKa/view?usp=sharing

Thanks Ken 

 

 

[Ken82]

M106 data which shows the issue more prominently-

Green     https://drive.google.com/file/d/1ifWDZs4gw2P3qIhyCVC5NjSmzltYzOUP/view?usp=sharing

Blue     https://drive.google.com/file/d/1hly5929pZCGT5bUBbMrxJmfhtZO3UD29/view?usp=sharing

Red      https://drive.google.com/file/d/1lrZdnf7eeJ_-hjmiT8YPCAfP-NLX4mg9/view?usp=sharing

 

[vlaiv]

I think this is alignment issue. In fact, I think it is very specific kind of issue that I'll try to explain.

First let's see if we can see what is happening in each corner. I took three corners to analyze. In first corner, we have this:

This is animation of R, G and B frames, Star positions look like pretty good match, there is no much shift between them, and if we measure centroid on a single star - this is what we get:

Star center is the same to first decimal place (roughly - there is difference in 0.1 pixel at most between star centers), so it is very good match.

Now let's see what happens in opposite corner:

Don't know if you can see this but here there is a bit more "wobble" in star positions between R, G and B. To actually measure it - let's do the same and select star and do centroid:

Ok, now we start to see that error in position is no longer 0.1 - it is larger, in fact, between R and G it is about 0.3 in X and almost the same in Y direction - total position being offset by almost 0.5px. Red and Green are closer and Blue is far from Red.

Now we need to see the third corner and see what the situation is there like, again, animation:

Here I see quite a bit of wobble in vertical direction. Let's again check actual numbers:

Interestingly - error in X is again 0.1 but error in Y is again being 0.3.

Do we see a pattern here? First corner both axis 0.1 error, second, diagonal corner both errors 0.3, third corner Y error 0.3 and X error 0.1. I wonder if fourth corner will show X error to be 0.3 and Y error to be around 0.1?

Interesting - this time error is not 0.3, it is 0.7 and it is in X axis. Y axis has same 0.1 error.

This means that Blue channel is more zoomed in, but how can this be?

We are dealing here with large sensor and relatively short focal length - wide field image.

Two things happen when we have wide field image. Projection distortion starts to creep in.

Extreme example of this comes from wide angle lens:

Straight lines are no longer straight lines in the image because of this type of distortion. Since FOV is only a few degrees - it is not really visible by eye but can become a problem if there is slight misalignment between images and you try to align them without first correcting for this distortion.

Second thing that happens to produce different level of magnification in wider field image is atmospheric distortion.

Atmosphere bends light. It particularly bends blue light (shorter wavelengths). This effect is more evident close to horizon than up high towards zenith.

If blue channel was shot when target was lower in the sky - then "bottom" part of the frame could be influenced more by this bending of the light - thus creating "zoom" effect for blue channel.

I guess that this could be fixed by applying different alignment model - one that also allows scaling rather than pure rigid transform. If wide angle distortion is dominant - it should be corrected (like lens distortion correction).

 

To see if this is viable option - check times when you shot each channel - and see if blue was last channel to be shot on particular night and also was the target nearest to the horizon. If M106 data shows this effect more - it could be that it was closer to the horizon at the time blue channel was shot?

 

[Ken82]

13 hours ago, vlaiv said:

To see if this is viable option - check times when you shot each channel - and see if blue was last channel to be shot on particular night and also was the target nearest to the horizon. If M106 data shows this effect more - it could be that it was closer to the horizon at the time blue channel was shot?

 

Thanks Vlad, very helpful as always!
 

I must say I find it hard to look at all the comments I get on this issue from different sources and try to understand what is correct. From people on the pixinsight Facebook page telling me they were horrified by their fsq performance and having to send it back to France and pixinsight themselves telling me it’s CA. It’s not easy ! 
 

So I checked times I collected M106 data. I did RGB in sequence so R being first. Started imaging R at about 11pm which was pointing exactly 1 hour from zenith (so atmosphere conditions should be minimal here)

M106 then past the meridian so I flipped and did the G images and finally B. The blue images were started around 1ish so again pointing only 1hr out from zenith and not really any lower than the R images. (So again atmosphere conditions should be minimal). 
 

In terms of the distortion you mention-  to limit this I’ve been doing a large crop on the images before alignment and I also enabling  distortion In star alignment pixinsight. (I’ll ask the developers if there is anything further I can do here to get a better alignment)
 

Since this issue at the moment always appears to be to the right hand side on all stars and different targets, I’m wondering if rotating the camera will rotate the issue on the star ? Would that tell us anything ? Do you take anything from the M106 timings ? 
 

thanks ken 

 

 

 

[fireballxl5]

I have a dual rig setup for AP with an 85mm scope for RGB and 150mm scope for lum. When combining these stacks in PI I usually have similar problems though more pronounced. In these cases I transfer the part processed stacks to MaxImDL to create the combined LRGB, going back to PI to finish off. I'd try a different app for RGB combination to see if it's just a limitation of PI. Having said this, mine may just be a problem with my lack of PI knowledge/experience.

[Ken82]

1 hour ago, fireballxl5 said:

I have a dual rig setup for AP with an 85mm scope for RGB and 150mm scope for lum. When combining these stacks in PI I usually have similar problems though more pronounced. In these cases I transfer the part processed stacks to MaxImDL to create the combined LRGB, going back to PI to finish off. I'd try a different app for RGB combination to see if it's just a limitation of PI. Having said this, mine may just be a problem with my lack of PI knowledge/experience.

It would be awesome if someone could demonstrate this on my M106 with different software. I only have pixinsight at this moment but happy to branch out if necessary. 

[vlaiv]

2 hours ago, Ken82 said:

Since this issue at the moment always appears to be to the right hand side on all stars and different targets, I’m wondering if rotating the camera will rotate the issue on the star ? Would that tell us anything ? Do you take anything from the M106 timings ? 

Not sure - I could not work with M106 data as it is in PixInsigth format and not fits like first set of data.

I did try different alignment process on first set of data and it corrected some of the issues. I checked star R, G and B centroid in two opposite corners and both shown 0.1px error.

I'll redo it and post results.

Maybe I could do the same for M106 if you post fits? Do you have luminance for these images?

[Ken82]

Sorry didn't realise i had posted pixinsight files. These are Fits files-

 

Red  https://drive.google.com/file/d/1eebbJxvJxjqIWxuUshnRKC_-xLWqvDl0/view?usp=sharing

Green    https://drive.google.com/file/d/1W84HFeholxHpN_P-2bkoypdGTi_cVi3h/view?usp=sharing

Blue    https://drive.google.com/file/d/1R-E1EoS3oL0DCuwKLqgGFd1Y38U44mP0/view?usp=sharing

Luminance    https://drive.google.com/file/d/1lVgrxNQxXH-Wp9tvM5kuvGY93xcaOJHO/view?usp=sharing

I've done a 70% crop on images as i did in my process to remove some distortion before aligning the RGB 

 

 

Edited May 28, 2020 by Ken82

[ollypenrice]

I've been speaking to Ken via PM and don't think the primary problem is channel alignment. I suspect that alignment is a secondary problem arising from lens distortions inherent to the data. I aligned the channels in Registar which, as well as aligning images, will resize them to fit a selected channel. I chose red as my reference channel because, though common sense might suggest green (mid spectrum) I've found that red is the most effective reference channel when problems occur. Here are two upwardly resampled crops from the centre of the image. I can see no reason to doubt  Registar's alignment and I don't see anything much wrong with them, though there are some probably unrelated pixel oddities.

However, the stars in the corners show classic elongation as if they were all part of a large circle. This is something we see with incorrect chip to flattener distances and other lens issues. I'll just post two diagonally opposed corners here because I sent all four to Ken and the same patter is sustained, theough the bottom left corner is much less severe.

 

I expect this pattern to be clear in the luminance as well but I've yet to download and stretch this. I'll do this later.

The distortion seems to be called sagittal astigmatism according to this website. https://www.lonelyspeck.com/a-practical-guide-to-lens-aberrations-and-the-lonely-speck-aberration-test/

Anyway I'll have a look at the lum.

Olly

 

 

[RayD]

Unfortunately it isn't unheard of for this kind problem to exist on an FSQ106.  I had to send mine off to Tak Europe (France) which was a bit painful, but it came back about 4 weeks later absolutely perfect.  I use mine with a APS-H sensor, which is a reasonable size, so it did show up the problems in the corners pretty clearly.

Edited May 28, 2020 by RayD

[RayD]

Is this using the SestoSenso to focus, Ken?

[vlaiv]

48 minutes ago, ollypenrice said:

I've been speaking to Ken via PM and don't think the primary problem is channel alignment

You have to take into account that same algorithm stacked all three channels into their respective stacks - and therefore aligned subs for each color while stacking.

Now - each sub was not taken at the same time and atmospheric impact could be possible between first and last sub.

Look at this crop and upsample (nearest neighbor - we can see pixel squares) and animated gif made by "blinking" three channels:

We can see distortion that you are talking about - all stars change shape slightly - being round or elongated, but large stars change position, and smaller stars stay in the same place???

How can that be if it is down to optics?

[Ken82]

41 minutes ago, RayD said:

Is this using the SestoSenso to focus, Ken?

Yea it’s using the sesto senso Ray. The fsq 106 focuser wouldn’t hold the camera without using it. In effect the focuser Is acting like a break. 
 

I had actually sent it back to FLO for it to be adjusted. I mean the camera and filter wheel are only just over 2kg so not much. A bit disappointing to be honest. 
 

so that was my first issue with this scope-
 

[vlaiv]

In the meantime ...

Ken why don't you try Deep Sky Stacker to see if it will do the same? As far as I know - it has rather "bendy" alignment model (sometimes it can really warp an image if it gets the stars wrong - so I guess it will try its best to compensate for any distortion made by lens or atmosphere).

[RayD]

Yes the smooth running of a focuser can be a bit of a problem for imaging.  I had exactly the same issue on the Televue APO I was using and had to use the focuser lock for anything above 60 degrees.  The electronic focuser do act as a brake which solves the problem.

I see yours is brand new so I wouldn't expect to see optical issues on it, but they definitely aren't unheard of.  Is that a ASI1600?

[Ken82]

4 minutes ago, RayD said:

Yes the smooth running of a focuser can be a bit of a problem for imaging.  I had exactly the same issue on the Televue APO I was using and had to use the focuser lock for anything above 60 degrees.  The electronic focuser do act as a brake which solves the problem.

I see yours is brand new so I wouldn't expect to see optical issues on it, but they definitely aren't unheard of.  Is that a ASI1600?

It’s Asi 6200m and filter wheel 👍

[ollypenrice]

28 minutes ago, vlaiv said:

You have to take into account that same algorithm stacked all three channels into their respective stacks - and therefore aligned subs for each color while stacking.

Now - each sub was not taken at the same time and atmospheric impact could be possible between first and last sub.

Look at this crop and upsample (nearest neighbor - we can see pixel squares) and animated gif made by "blinking" three channels:

We can see distortion that you are talking about - all stars change shape slightly - being round or elongated, but large stars change position, and smaller stars stay in the same place???

How can that be if it is down to optics?

Potentially it depends on the colour of the star, perhaps? My attempt to download the Lum failed so I'll try again.

Olly

[RayD]

1 minute ago, Ken82 said:

It’s Asi 6200m and filter wheel 👍

Nice!

I'll have a look at your images with APP later to see if that does anything different.  PI is generally pretty good though, so I'll run it the way I do mine in that.  

Do you do a manual process or do you run the batch pre-process script, and are you using default settings?

[vlaiv]

1 minute ago, ollypenrice said:

Potentially it depends on the colour of the star, perhaps? My attempt to download the Lum failed so I'll try again.

Not sure if it does - since all stars show in each channel - I would expect that they have enough signal in each band.

[ollypenrice]

1 minute ago, vlaiv said:

Not sure if it does - since all stars show in each channel - I would expect that they have enough signal in each band.

But if there is an optical aberration it might well vary with wavelength? In other words it might be chromatic to some extent. I'm still trying to download the L. That will decide it, I think.

Olly

[vlaiv]

2 minutes ago, ollypenrice said:

But if there is an optical aberration it might well vary with wavelength? In other words it might be chromatic to some extent. I'm still trying to download the L. That will decide it, I think.

Olly

I see similar optical aberrations in all stars in the image per channel. Red is most round. Green shows sagittal astigmatism. Blue shows inward coma like tail. I'm not sure if that is aberration or stacking artifact - maybe stars moved due to atmospheric influence between first and last frame and that tail is sigma clip artifact - or it could be genuine thing.

M106 Luminance top right corner:

Top left corner:

Bottom left corner:

Bottom right corner:

Same as green in above animation that I made - only significant thing that I see is sagittal astigmatism

[ollypenrice]

OK, here's the stretched lum, top left. The other corners show the same sagittal astigmatism effects. I think that they may be exaggerated by the wavelength (blue being the worst) but that might be expected, maybe?

I think Takahashi should speak out on this. The chip we are dealing with does not reach out into the limits of their claimed corrected circle. Anyone trying to use a 36x36mm chip in this scope would be doomed.

When the stars are distorted differently at different wavelengths the software will struggle to align channels. I think Registar did as well as could be expected but, when we see all visible wavelengths captured together, as here, the distortions are still present and have nothing to do with channel alignment. 

The scope is not working to spec in my opinion.

Olly

[Ken82]

1 hour ago, ollypenrice said:

OK, here's the stretched lum, top left. The other corners show the same sagittal astigmatism effects. I think that they may be exaggerated by the wavelength (blue being the worst) but that might be expected, maybe?

I think Takahashi should speak out on this. The chip we are dealing with does not reach out into the limits of their claimed corrected circle. Anyone trying to use a 36x36mm chip in this scope would be doomed.

When the stars are distorted differently at different wavelengths the software will struggle to align channels. I think Registar did as well as could be expected but, when we see all visible wavelengths captured together, as here, the distortions are still present and have nothing to do with channel alignment. 

The scope is not working to spec in my opinion.

Olly

That’s already a 70% crop so the distortion is much worse further out! 
 

In terms of the astigmatism Vlad mentions at some targets it is purely sagittal astigmatism but then depending on where the scope is pointing I get tangential astigmatism in some corners. I think the draw tube may be sagging? 
  
I also use a 36mm sensor with my fsq 85 (Without additional flattener) and I’d say the astigmatism on this Is similar but then it’s corrected field is much smaller!  
 

 

Edited May 28, 2020 by Ken82

[ollypenrice]

25 minutes ago, Ken82 said:

That’s already a 70% crop so the distortion is much worse further out! 
 

In terms of the astigmatism Vlad mentions at some targets it is purely sagittal astigmatism but then depending on where the scope is pointing I get tangential astigmatism in some corners. I think the draw tube may be sagging? 
  
I also use a 36mm sensor with my fsq 85 (Without additional flattener) and I’d say the astigmatism on this Is similar but then it’s corrected field is much smaller!  
 

 

Here's a corner crop in luminance from my elderly FSQ106N on a 45mm diagonal chip resized up by the same amount as I did yours. (Atik 11000 camera.) It isn't perfect but I think it's significantly better. I never have any channel alignment issues arising in RGB.

Olly

Edited May 28, 2020 by ollypenrice

[fireballxl5]

potential issues with the optics aside, here's a 100% crop of the LRGB combined image using MaxImDL v6.09 for comparison.

HTH, Andy