Adding "L" Data To RGB Data in Pixinsight

Matching the intensity of L and RGB when combining LRGB non-linearly is, at least in my opinion, in the top 3 most difficult parts of processing. There is a simpler way though. L can be combined with an effectively perfect intensity match while linear using a very simple and predictable procedure:

1. Basic processing on both L and RGB (Color calibration necessary!)

2. Extract the average of the color calibrated RGB

3. Linear fit L to that average

4. Multiply RGB by L and divide by the median of L to preserve the original brightness - RGB * L / med(L)

If done correctly, you won’t need to worry about stretching both L and RGB to the same intensity, washed out colors, or other issues with non-linear combination.

As to how processed both images need to be processed before combining, it’s kinda preference. I would definitely remove gradients and calibrate the colors. I also like to blurx both images separately. Same with denoise - RGB at 100% and L to taste. But I prefer to extract stars after combination, so there aren’t any issues with galaxies getting extracted from RGB but not from L or vice versa.

Jim Raskett · Jan 4, 2026, 03:23 PM

Hi Mikolaj,

Thank you very much for the response! Sounds like a fairly simply process, but I want to be sure what you mean by "Extract the average of the color calibrated RGB". It is probably something quite simple, but possibly I am over thinking what it means. What exactly am I extracting and what do I do with the extracted average?

Thank you! 
Jim

It’s just the average of the RGB channels avg($T[0], $T[1], $T[2]).
It is indeed simple but tbf I phrased it quite poorly.

Jim Raskett · Jan 4, 2026, 04:54 PM

Hi Again Mikolaj,

I appreciate the response and the extraction PM worked perfectly. 

Having some Incompatible image geometry issues running LinearFit on the L image. I tried registering the two images and applying the same DynamicCrop tp both and still getting the same error.  
I am sure I am just being humbled by PI!

Much thanks,
Jim

Hi,

The image you’re creating with pixelmath is probably a color image. By default, pixelmath makes an image with the same amount of channels as the target image, so here it’s three. You just need to set it to create a new grayscale image instead, then it’ll work. Otherwise, you can’t linear fit a monochrome image to a, technically, color image.

Jim Raskett · Jan 4, 2026, 05:47 PM

@Mikołaj Wadowski

I just figured that out! I converted the extracted to grayscale and LF worked perfectly. Below is the result. The Extracted RGB image on the left and the "L" image on the right that has had LF applied. The brightness looks quite different. Does this look like the expected result?

I apologize for so many questions. I didn't want for you to have to spend this much time on responding to my post!

Hm, yeah that doesn’t look right, they should be near identical in brightness.

I assume they have the same STF applied? It looks like they do but it’s better to rule out the simplest solutions first :)

If they do, try lowering the ‘reject high’ parameter in LF. If that doesn’t work, I see a dark halo around the galaxy in your RGB data, likely a gradient extraction artifact. Try redoing gradient extraction, this time being more careful not to extract actual signal. Lum on the other hand looks perfectly fine if you need a reference of what to aim for.

My guess is that either the dark halo or different clipping points might be confusing LinearFit (the latter being less likely but easier to test).

And don’t worry about the questions! These are perfectly valid. If you need any further help, feel free to ask.

Jim Raskett · Jan 4, 2026 at 01:55 PM

I have been struggling with my first LRGB dataset of M33 for several months now. I have been an OSC shooter for many years and have enjoyed OSC processing. I added a Minicam8 mono to my kit awhile back and have had tons of fun with mono processing, but only with narrowband data.

It seems to me that my biggest issue is how to incorporate the Luminance data into the RGB image in Pixinsight. I have seen several videos on YouTube that work somewhat for me, but they are all quite different in approach. I have been working with the Pixinsight LRBG Composition series and have just gotten to the point of adjusting the brightness/contrast of the non-linear processed L image to closely match the brightness/contrast of the partially processed non-linear RGB image. Then adding the L image to the RGB image with ChannelCombination in the CIE L*c*h* Color Space. This can result in a nice brightness/detail enhancement to the RGB image but only if I adjust the L image just right. Determining “Just right” is one of my issues, but more important, is this the best way to go about adding L to RGB?

Another question that I have is just how far to bring both the L image and the RGB image in processing before blending?

I have read that using the ImageBlend script offers another method and a better choice to combine L with RGB. I have looked at it, but would need some work to fully understand it.

I am looking for some basic information like combination methods and how/when blending plays into workflow.

Thanks a bunch for any tip in direction that you can give me.

If you look through their educational videos, Pixinsight has a great video about how to combine Lum with RGB. I also talked about this subject in my TAIC talk, “Galaxy Studio” that you can watch here: https://www.youtube.com/watch?v=1XBon7x6kio.

John

Yes, that’s the PI video that I was referring to. If you wade through the whole thing, it is pretty good. The method that I outline in my TAIC presentation is very similar. I think that if you can find the time, watch both videos (mine and the one from PI). Sometimes setting the speed at 1.25x helps to speed it up a bit and you can always slow down if you miss something.

Good luck with it!!

John

Marco Montella:
I read the available resources (including John’s excellent lecture at TAIC!), and it seems that LRGB merging was coded up and intended to work strictly on non linear inputs.

On the other hand, it’s not clear to me why the linear method should not be used. I have been following this pipeline for months, with (at least apparent) success:

RGB: gradient correction, color calibration, mild deconvolution, star removal

Lum: gradient correction, mild deconvolution, star removal

Linear Fit of Lum to RGB_L component > LRGB Combination

Interestingly, LF results are substantially better if the stars are removed from both reference and target. I imagine that (similar to SPCC), saturated pixels can throw off linearity and generate skewed results such as those shown by Jim Rasckett.

Marco Montella:
I read the available resources (including John’s excellent lecture at TAIC!), and it seems that LRGB merging was coded up and intended to work strictly on non linear inputs.

On the other hand, it’s not clear to me why the linear method should not be used. I have been following this pipeline for months, with (at least apparent) success:

RGB: gradient correction, color calibration, mild deconvolution, star removal

Lum: gradient correction, mild deconvolution, star removal

Linear Fit of Lum to RGB_L component > LRGB Combination

Interestingly, LF results are substantially better if the stars are removed from both reference and target. I imagine that (similar to SPCC), saturated pixels can throw off linearity and generate skewed results such as those shown by Jim Rasckett.

Craig Towell:
Not sure if it’s of any use but I follow the method whereby a stretched and processed Luminance image is combined with a stretched and processed RGB image using channel combination, but I use the CIE L*a*b colour space. I have the L box checked and select the luminance image for that. The a and b boxes are unchecked, and then I drag the instance onto the RGB image, which is then replaced with the LRGB image.

Before doing this I stretch the two images by eye to the same intensity, using the histogram in histogramtransformation as a guide.

Before combining I usually also give the RGB image a good blur with convolution tool, to really smooth it out. And also do some selective desaturation of the background if required if it is a bit colourful (using inverted luminance mask and the saturation tool)

Hi Jim,

When I’m having a hard time blending the RGB and Lum, I do find the ImageBlend approach can help. The main advantage is that it allows you to interactively tweak the blackpoints/midtones of each of the two layers to help you get a better match. A couple of caveats:

1) You should try and get the Lum/RGB matching as closely as possible as described in the Pixinsight videos.

2) When you’re tweaking the black levels, you do need to make sure you’re not clipping the layer.

—Steve

Marco Montella · Jan 5, 2026, 08:27 AM

Linear Fit of Lum to RGB_L component > LRGB Combination

Can u explain in detail this step ? Thank u :)

And for confirmation, the strech after the LRGB right ?