Hello all
I appreciate this is an old discussion but I'd be very interested to hear views here and Robin's views on it. With modern CMOS sensors, and crucially the advent of AI deconvolution tools such as BlurXterminator, is Lum still needed as it was?
In terms of processing, many will treat Lum differently to RGB and then combine them later. There is no reason that a synthetic Lum can't be created from the combined RGB frames and processed separately.
I think much of the thinking in AP seems to stem from the CCD days. Now we have AI powered deconvolution and noise reduction and very sensitive CMOS sensors.
I am going to be doing an actual test on my remote scope on this once it is set up (e.g. 6 hours RGB vs 3 hours L 1 hour each RGB), but in the mean time I would appreciate your opinions!
LRGB vs RGB with modern CMOS sensors.
-
admin
- Site Admin
- Posts: 13362
- Joined: Sat Feb 11, 2017 3:52 pm
- Location: Vale of the White Horse, UK
- Contact:
Re: LRGB vs RGB with modern CMOS sensors.
Hi,
to do a fair comparison I suspect you should make a synthetic luminance in both cases... In the RGB case it is just from the 3 channels, in the LRGB it will be from the 3 hours of L plus the 3*1 hour colour channel exposures. The LRGB will be based on twice as many photons in total, so the SNR will be 40% better in that case before noise reduction.
Now, AI noise reduction will create a low noise image for you - partly based on the image that you feed it and partly based on images that formed the training data for the AI algorithms. The more noise there is, the more the fine detail in the processed image will be 'synthesized' based on the training data. From the point of view of pictures that look nice, it probably doesn't matter, but if you go too far down that road you get to the 'Samsung phone photo of the moon' problem where they were caught pasting a high res image of the moon into any image that looked like it had a blurry/out of focus moon in it (see https://www.reddit.com/r/Android/commen ... _and_here/ and https://imgur.com/ULVX933)
cheers,
Robin
to do a fair comparison I suspect you should make a synthetic luminance in both cases... In the RGB case it is just from the 3 channels, in the LRGB it will be from the 3 hours of L plus the 3*1 hour colour channel exposures. The LRGB will be based on twice as many photons in total, so the SNR will be 40% better in that case before noise reduction.
Now, AI noise reduction will create a low noise image for you - partly based on the image that you feed it and partly based on images that formed the training data for the AI algorithms. The more noise there is, the more the fine detail in the processed image will be 'synthesized' based on the training data. From the point of view of pictures that look nice, it probably doesn't matter, but if you go too far down that road you get to the 'Samsung phone photo of the moon' problem where they were caught pasting a high res image of the moon into any image that looked like it had a blurry/out of focus moon in it (see https://www.reddit.com/r/Android/commen ... _and_here/ and https://imgur.com/ULVX933)
cheers,
Robin