Best way to combine binned images?

[bitli]

Hi,
This may have been discussed already, but I cannot find the post.

I have L images and 2xbinned  R, G and B. Currently I have a script that register the L ones, then register all R,G,B on one of them (star alignment).
Then I integrate them (L, R G and B separetely), I rescale the R, G and B by 200 % and register the resulting ones with the integrated L.
And finally I do an LRGB combination. The dbe, and so on.

Is this the best order? Would it make sense to align the R G and B on a half size integrated L instead of using an arbitrary R? Would it make sense to combine the align RGB and then resize them and add L? Does it make any difference?

I understand that taking non binned would be preferable, but time does matter (I use lightbucket.com).

-- bitli

- bitli

[Nocturnal]

I think your decision centers around what size you want your final image to be. As I understand it the reason color frames are binned is that the detail in the L frames and we can afford to have the color data at half that resolution due to the eye being more sensitive to brightness than color variation.

So I think your steps are fine except that the 2x upscale isn't needed. Registration should take care of that. It won't hurt but it would skip a step. Give it a try and see if it does indeed work

[Carlos Milovic]

IMHO, you should upsample your images, and then register using L as reference. Making registration twice (with respect to R, or other color channel, and then to L) will degrade the image more than a simple registration.
Maybe it is not necessary to previously upsample the data, and just use L as reference. This will speed up things a bit.
(To those who may want to know the reasoning: there are experiments where working with artificially upsampled data yields in fact better results than with original data. This is very much the same here... You would get a slight better resolution or SNR integrating your upsampled and registered images than stay with the binned size until the end).

[Andres.Pozo]

IMHO, you should upsample your images, and then register using L as reference. Making registration twice (with respect to R, or other color channel, and then to L) will degrade the image more than a simple registration.

You don't need to first upsample and then register. As Nocturnal said in a previous post, if you register the R-G-B frames using L as reference, the results are upsampled. I would think that the results are better if done in only one step than two.

[Carlos Milovic]

Yes. I got sidetracked thinking on different interpolation algorithms. One for upsampling, and then the one for registration.

[bitli]

Indeed it seems to work fine when registesring different sized image, to my big surprises. It attemps 2 ransac, and results in
Homographic Projection / Bicubic Spline Interpolation, c=0.30: 100%


When I think the time I took to split the images by size... I should have read the doc :-)

thanks for your help
-- bitl