Color Calobration - Incorrect Color Outcome

[dpaul]

Hi

I'm having a frustrating time with color calibration today. This is using a one-shot color camera (Atik Horizon) and LRGB filters.

I've only recently started LRGB imaging and as can be seen from posts I did in the gallery section last week, in general my first attempts have shown reasonably good color balance.

That was until last night!

I was taking light frames of NGC 3718 and the resulting combined RGB image is too green. This galaxy should be more blue with a gold/yellow bar across it but that bar is green and not much evidence of blue.  I have calibrated each light frame with darks, bias and integrated them into separate R,G, B and L master ''linear'' images. I've then done star reduction using MT and deconvolution - all with use of star mask and luminance masks. Then I used channel combination to combine the ''still linear'' R,G, and B.

The resulting background when stretched was green (but I've seen that before in tutorials-I'm assuming that's ok?). Next step was to color calibrate so I first did background neutralisation which removed the green, then I used color calibration with two preview images (background without stars) and a bright star for whilte balance. The result was too green, nothing like images I've seen on the internet of this galaxy. I also tried photometric color calibration which at first worked then when I tried again later it refused to - but this still gave a similar result.

After stretching the RGB image with HT and increasing saturation with curves, I still cannot see anything but a green bar across this galaxy.

Must be doing something wrong?

David

[bonz0]

This is using a one-shot color camera (Atik Horizon) and LRGB filters.

First of all, why are you using a one-shot color camera AND LRGB filters?! That doesn't make any sense at all. You already have a color camera, why put filter in front of it?
People use monochrome cameras with LRGB filters because of the extra bit depth they provide. Other people use one-shot color cameras because of the ease of use. Using LRGB filters with a one-shot color sounds like the worst of both worlds.

Anyway, if we ignore that,

Then I used channel combination to combine the ''still linear'' R,G, and B.

Try using LinearFit on the R, G, and B frames before combining the channels. After that you shouldn't need to perform ColorCalibration.

If you are doing ColorCalibration:

I used color calibration with two preview images (background without stars) and a bright star for whilte balance.

Why are you using a single star for white balance? Do you know that that particular star is perfectly white? Use <Target Image> instead of a single star for white reference if you're doing ColorCalibration.

And if you still get green, then try using SCNR.

[pfile]

did i read your steps right? did you do color calibration after stretching the images?

rob

[dpaul]

Hi Rob,

Sorry, I mixed up two issues (I have a separate issue with getting color after debayering on a one-shot color camera - see other post).
I meant to say it was mono camera - stupid me!

I've now read about linear fit and can see how it can match the R, G, and B channels before combining. However, non of the tutorials I've read say ''where to get the reference image from''. I've no idea what to do at all here, for eaxmple something from the internet that ''looks right'' for that same object.

Regarding color calibration, what would you use as the white reference instead of 1 star? I've seen people using a preview of a galaxy but what is that galaxy is muti-coloured?

Thanks

David

[RickS]

For LinearFit you usually just use one of the colours as the reference and apply to the other two.

Wrt the white reference for ColorCalibration the default is to use all the stars in the field (assuming that on average they'll be white.)  As you mentioned, a whole galaxy generally works as a reference too - most galaxies are a mix of new, blue stars and older red ones.  Once again you get an overall reference that is roughly white.

Using a single star is pretty risky unless you know it is really white, e.g. a G2V star.

The new PhotometricColorCalibration process is my preferred method of colour calibration.  It's a bit more work but worth it.

Cheers,
Rick.

[dpaul]

Thanks Rick,

With Linear fit, how do you physically select one color and compare it against the others?

The only option is to pick ''one'' reference image and apply to the ''one active image'' so I still don't know how to do this?

I also like Photometric Color calibration but its not easy to get it to work all the time. I've even had the same data working / not working on different days.

Thanks

David

[pfile]

linearfit can probably get you into the ballpark in terms of color, but in terms of the math it does the same thing that color calibration does, albeit a little more blindly. IMO there's no requirement to do LinearFit on RGB images if you are going to do color calibration - in essence the CC step is just going to undo whatever LF did.

here is where i think LF is applicable:

1) when doing RGB combines of narrowband data. since color calibration is 'meaningless' with SHO blends, CC may not be applicable (i say may, because people do use PhotometricColorCalibration "off label" with SHO data and it apparently works, but the results are physically meaningless of course. and one way to balance the channels in an SHO image is to use ColorCalibration over the entire image, but this is sort of equivalent to using LinearFit...)

2) when doing DBE on RGB data where the channels have wildly different brightness (which is pretty common of course). because DBE has only 1 thresholding control, in order to pick up pixels in the brightest channel you may need to set the threshold control so high that the other two channels are oversampled. in this case it helps to have the channels aligned in brightness before you start with DBE.

3) when making mosaics, to get the brightness of each pane to similar levels before assembling the panes. however, LF needs to be run on registered images, which of course mosaic panes are not. you need to run LF on the intersection of the panes, which was a hassle until David Ault wrote a script called DNALinearFit which does the intersection calculation automatically.

as far as your question about which image to pick... just pick one channel as the reference and apply LF to the other two channel images. it's all relative so to speak so it does not matter which image you choose as the reference as long as you use it on both of the other two images. not to confuse the point, but in fact if you chose G as your reference and applied it to B, then chose B as the reference and applied it to R, the result would probably be much the same as using G as the reference for B and R since after linear fitting B, it's statistics should be comparable to G.

rob

[RickS]

Hi David,

With Linear fit, how do you physically select one color and compare it against the others?

The only option is to pick ''one'' reference image and apply to the ''one active image'' so I still don't know how to do this?

If Rob's description wasn't enough... you select any one of the colour masters as the reference using the "Reference Image" box on the LinearFit dialog.  Then you drag the triangular new instance icon from the LinearFit dialog and drop it on each of the other colour masters.

I agree with Rob that this isn't an accurate method of colour calibration but it can produce reasonable results.  A similar method is to use an unlinked STF stretch.

I also like Photometric Color calibration but its not easy to get it to work all the time. I've even had the same data working / not working on different days.

I use it a lot and have had no problems.  I take my registration reference sub, use the ImageSolver script to plate solve it and then use the "Acquire from image" button on PCC.  I think that doing the plate solve separately is helpful.  Most people that have problems with PCC seem to be having problems with the plate solve, not with the colour calibration.

Cheers,
Rick.

[dpaul]

Hi Rick,

I checked the filter wheel just to make sure I had the right filters in the right positions - which was correct.
I've also tried linear fit (and nothing else), then tried color calibration (without linear fit) of each R,G and B integrated frames and finally tried Photometric Color calibration (without linear fit). The attached two images are the final results after merging with Lum and then stretching. V1 is color calibration, V2 is PCC.

As you can see, the results are quite green, the bar across the main galaxy in internet image sis always golden, not green. I tried SCNR which eliminated the green but also most of the color in general.

Thanks for everyone's help about better understanding of linear fit and when to do / no to do, etc. There is however still some fundamental issue with the color management. Its odd because other LRGB results seem quite good until now??

Thanks

David

[RickS]

That's pretty odd, David.  Are you able to make the R/G/B masters available on Dropbox or similar for download?  I'd be happy to take a quick look at them.

Cheers,
Rick.

[dpaul]

Hi Rick,

It doesn't sound very secure giving public info about dropbox, so I'm concerned about doing this 'on a forum'.

I did try one more time - this time I calibrated and registered all frames in one go. Previously I'd done the R,G<B and L separately then star aligned before RGB intregration.
The results are bit different, probably because I color calibrated slightly different - attached is this version.
Not so green.

Maybe I did mix up frames when saving the data from the camera? I did have a break when doing the blue due to temporary cloud cover.

Lets see how my next session goes.

thanks

David

[RickS]

No problem, David.  Could definitely be a result of mixing up frames.

Cheers,
Rick.

[dpaul]

Hello Rick

I re-imaged NGC 3718 last night and did a quick Pixinsight post processing today and the colors are now ok-
So I must have mixed up some frames the previous time. Last nights result were not ideal due to incorrect tracking on my equatorial platform (didn't align as well I should have done) but will try again another night - one of my favourite objects!

Thanks

David

[RickS]

Glad you got it figured out, David!

[Geoff]

With Linear fit, how do you physically select one color and compare it against the others?

The only option is to pick ''one'' reference image and apply to the ''one active image'' so I still don't know how to do this?

David

I look at the brightness of the three images using the statistics tool, then pick the middle one to use as the reference--usually it turns out to be the G.