Registration of 2x2 RGB to Lum 1x1

[stevek]

Hi guys, regarding registering my RGB 2x2 subs to a luminance 1x1 master.

I learned a technique from Adam Block (learned many techniques from Adam actually) . I use the Integer Resample process to downscale a copy of the master registration luminance 1x1 sub (that the entire data set is registered to) and downsample this to 2x2.   I then register the 2x2 RGB subs to the newly downsampled 2x2 copy of the master luminance frame. This gets all the RGB subs aligned with the 1x1 luminance (via the 2x2 downsampled copy).  Cool.

My questions:
1.  After registration should I then upsample all the newly registered 2x2 RGB subs up to 1x1 before image integration?  Or should I leave as 2x2 and let LRGB combination take care of upscaling the RGB after processing of the resulting chrominance is complete???

2.  About the master luminance registration frame.  Should this be the original sub off the camera or should it be a frame that has been calibrated and cosmetically corrected prior to being used as a registration reference?  I can't see that it matters but wondered if anyone had a view on that?

Thanks, Steve

[ngc1535]

Hi Steve,

In our last conversation it was concerned with flats (I thought). Now you have a different resample question.
ImageRegistration through StarAlignment does not require the images to be the same size. I typically produce combined R,G,B images that I register with a larger luminance (reference) without resampling.

1. So I think the answer to this is no. You can combine your 2x2 images...and then you will have 4 images. 3 images (RGB) that are 2x2 and one (luminance) that is 1x1. Just align these 4 images with Dynamic StarAlignment with the luminance as the reference.

2. Based on my suggestion above- why not align with your combined luminance (that is calibrated, cosmetically corrected, and has the best S/N)? I think this is the way to go.

-adam

[stevek]

Thanks Adam!
Yes the previous question was about flats.

In one of your videos you downsample a 1x1 luminance to 2x2 and use that as the image registration reference for the 2x2 RGB's.....  It's that I was referring to.

Great idea about the master luminance as the RGB registration image reference!

[ngc1535]

Hi Steve,

You are right, I did do what you said... I was trying to be clever.

And it might be a good point of discussion by more knowledgeable people ... This whole thing is obviated if all of the data is taken at the same scale (ie 1x1). However, if you have 1x1 Luminance and 2x2 RGB data- it is important not to upsample and register the 2x2 data before doing rejection (through ImageIntegration). The interpolation of pixels by the upsample and registration can smear small features and it make more difficult to reject the small features (satellite trails, cosmic rays...etc).

So this leaves two methods:

#1. You can register all of the 2x2 subframes within their color group. So for the RED data I pick a Red frame and register all of the Red data and then do Image Integration. The same for each of Green and Blue. Now these combined images are aligned with the 1x1 Luminance. But if you notice...this requires two registrations of the color data. In my mind it doesn't hurt anything because the color data doesn't need to be exactly the same resolution and making it a tiny bit fuzzier by registering twice is not important... but...

#2 What if you think that double registration of the color is an issue? Well... you could downsample the luminance reference (either a single frame or the combined final luminance if you already have it) and then align the RGB data. This would result in a single registration (and interpolation) of pixels. Since the interpolation smoothing will remain smaller since there is no upsample yet before ImageIntegration- the theory is that maybe this is better? The RGB combined images would later be upsampled after ImageIntegration.

I actually do not know the answer! I demonstrated #2 in my video because I liked the idea and wanted to demonstrate a variation. You have options.
The takeaway is that for any data, I do strongly believe you do not want to introduce large interpolation-smoothing before doing ImageIntegration. The two methods above keep the RGB in their 2x2 state for rejection (in ImageIntegration).

I do like method #2- but there was some context/background as to why I was showing it.

-adam