Details of LRGB Combination

[jerryyyyy]

Looks like I have progressed enough to start asking some questions about alternative practices in LRGB formation. 

Seems to me like the best first step is to do a LinearFit on all 4 images so that they are not jammed at the left side of the histogram. 

Then LRGBCombo versus ChannelCombination.  Both seem to get to approximately the same place, but LRGBCombo has the Chrominance noise reduction feature and star colors look a bit brighter.....

Then to add in L, either: extract L, a and b with ChannelExtraction and use ChannelCombination with the original L and a & b;  or use LRGBCombo to add in the L.  The later process allow different percentages of images to be added.

Thanks

[troypiggo]

I'm no expert, but my current workflow is:

- Combine the linear data RGB only (no L yet) using the LRGB combination tool with all default values;
- Background Neutralise and Color Calibration to the RGB image;
- Stretch to get non-linear RGB image that looks in the right ballpark (Histogram Stretch or Masked Stretch script);
- Use Channel Extraction to get the L* from LAB of the RGB image;
- Stretch the L non-linear to get something like the L* image;
- Linear fit the L* image to the L image;
- Re-insert the L* back into RGB LAB mode using Channel Combination;
- Now use LRGB Combination to add L to the RGB;
- Continue processing your LRGB combined image.

I was having a heck of a time combining my 1x1 binned L with 2x2 binned RGB until I read Juan mentioning the method above.  Main place I was going wrong was trying to combine the L and the RGB while linear data, but should be done after stretching.  The other trick is the Linear fit of the L* from RGB only with the L subs.  That works beautifully.

[papaf]

To the both of you, take a look at this video tutorial: http://pixinsight.com/videos/NGC1808LRGB-vperis/en.html
Troy, I used to stretch RGB and L to "look alike" like you do, but in thi tutorial, Vincent Paris explains a really clever method using ScreenTransferFunction which actually equals the mean values of the two images. It's awesome and the results when then combining L to RGB are really much better!

[pfile]

the STF thing is a very good idea, but sometimes your data can't support the stretch that STF applies by default. this is often the case for my light polluted skies. i suppose you can adjust the STF controls so that it does not stretch as hard, but sometimes it's easier just to do your stretches by hand.

[jerryyyyy]

To the both of you, take a look at this video tutorial: http://pixinsight.com/videos/NGC1808LRGB-vperis/en.html
Troy, I used to stretch RGB and L to "look alike" like you do, but in thi tutorial, Vincent Paris explains a really clever method using ScreenTransferFunction which actually equals the mean values of the two images. It's awesome and the results when then combining L to RGB are really much better!

Thanks for reminding me of this... I looked at this video when I started by it bears a second or third look. 

With the LRGB Combo I now notice that this all occurs in the non-linear mode and you have to stretch the images first.  This then could lead to some over saturation of the high end, which they show how to deal with using HDR wavelets on the RGB and L images before combining.  The other black science here is how much of each to apply.

I wonder if the alternative ChanelCombination method differs primarily in that it occurs in linear space.  I guess you deal with the highlighing by using the LinearFit function to all images before combination.  You have to work with the saturation on the combined image later after you would convert the whole thing to non-linear space. 

I am running both paths down from the same original 17h worth of data from M101. 

[troypiggo]

To the both of you, take a look at this video tutorial: http://pixinsight.com/videos/NGC1808LRGB-vperis/en.html
Troy, I used to stretch RGB and L to "look alike" like you do, but in thi tutorial, Vincent Paris explains a really clever method using ScreenTransferFunction which actually equals the mean values of the two images. It's awesome and the results when then combining L to RGB are really much better!

I was using the STF method previously, and it is a good starting point to show you something in the right ballpark, but I find the LinearFit method suits me better.  I also found the STF method stretching too much in one go and stars were blowing out.

[jerryyyyy]

To the both of you, take a look at this video tutorial: http://pixinsight.com/videos/NGC1808LRGB-vperis/en.html
Troy, I used to stretch RGB and L to "look alike" like you do, but in thi tutorial, Vincent Paris explains a really clever method using ScreenTransferFunction which actually equals the mean values of the two images. It's awesome and the results when then combining L to RGB are really much better!

I was using the STF method previously, and it is a good starting point to show you something in the right ballpark, but I find the LinearFit method suits me better.  I also found the STF method stretching too much in one go and stars were blowing out.

You know, I just noticed those blown out stars too.  I have been running the same data in both pipelines very carefully.  Color seems easier to adjust in the LRGB Combination, than in LinearFit but the LinearFix runs in linear mode so it seems that after all is combined then you can change to non-linear and use all the tools you need there.  Anyway, at least I have enough control of this beast to run them side by side.  I have to look at the stars... maybe there is a fix but I'd rather avoid the problem in the first place. 

[papaf]

STF can be tweaked to stretch less than the default. The big plus is that you're mathematically sure the median value of both images will be the same.

[jerryyyyy]

STF can be tweaked to stretch less than the default. The big plus is that you're mathematically sure the median value of both images will be the same.

Are you saying that using STF on both the RGB image and the L image and then using LRGB Combo is more assured to give you a level set of colors than ChannelCombination and LinerFit?  I can believe that.  Below is the same data as the above image processed down the ChannelCombination LinearFix pipeline.  The stars are smaller but I had to do a lot of stretching after converting the composite image to nonlinear with STF and I think the image suffered from that.

Thanks for bearing with me as I have only 1.5y in astroimaging and 2m with PI.

[pfile]

well, the thing is when you are doing the LRGB merge onto your RGB image, what you are doing is replacing the implicit L channel of the RGB with the L image. if the two images do not have similar histograms then the merged result is going to look funny, as the implicit L will be dimmer or brighter than the L image.

using STF to compute the HT for both images virtually ensures that the resultant images have very similar histograms, that's all.