Linear Process ordering

[MattJenko]

Morning All,

I am a recent convert to PI and have been watching lots of the excellent video tutorials out there and practicing a lot on my own datasets. I have noticed however, that the ordering of certain actions in people's workflow varies when it comes to doing linear processing, namely BackgroundNeutralization/DBE/ColourCalibration. Does the state of the data in the image dictate which of these processes should be performed first, or are they mutually exclusive enough that it does not matter?

My own view is that given what each of these processes do, I can see that each of these has an impact on the other, and if there are certain dependencies, it would be nice if someone could give me an idea of why certain processes should happen before others.

Many thanks
Matt

[jkmorse]

Matt,

Depends on whether you are using a DSLR or a mono CCD.  If the former, probably makes more sense to do the BN and CC first, then DBE.  If mono, I always do my DBE first on each stack, then run it again after doing a channel combination and BN/CC to clear up any residual issues in the color image background.

Hope that helps,

Jim

[MattJenko]

Thanks Jim,

So you process each RGB channel before combining? I did wonder if I should be processing the individual channels more before crunching them together, as most of the videos I have seen start with the RGBCombination as the first step. I'll give that a go. Apologies for the follow on question, but.... is there a theory behind the ordering, or is this something you have just found out yourself through experimentation? Not questioning the validity of the approach, just curious.

Thanks again
Matt

[jerryyyyy]

This question(s) has always bothered me too.  I have seen arguments on all sides.  One fact that I think should be clarified is whether or not you need BE if you use DBE.  I think DBE in its latest evolution is supposed to do the BE. 

On the one hand, DBE also works on each channel separately so there should be no reason to wait before combining them.  On the other hand, it seems to me you want to a LinearFit on the channels before combining them... seems that would make the CC easier... or does it introduce an artifact? 

Finally for maximum confusion, is there any difference between LRGBCombo and ChannelCombo for putting the RGB together.  For that matter, could you just use PixelMath? 

Should be a PI Bible somewhere... but then you would have interpretations..

[msmythers]

Just thought I'd throw my 2 cents worth in here. If you have varying color gradients in an image due to what ever, it would seem to me to make more since use DBE/ABE before color corrections. If you have perfect source data though I would think it would not matter which order was done. 

Also if your working with filtered channels individually the samples and settings for DBE/ABE should be the same for each channel. This way your keeping perfect calibration of the channels through these processes. Mathematically what you do to one channel you do to the other channels.

One interesting thing is the normalization function in DBE/ABE. If samples are selected properly so no stars or nebulosity are sampled then this seems to act the same as BackgroundNeutralization. This action does not happen at the beginning of theses tools but at the end. This to me would indicate that color correction is normally thought to happen after DBE/ABE.


You my thoughts.


Mike

[jkmorse]

Matt & Jerry,

I am with Mike on this one.  It really depends on the gradients.  Since I shoot from areas that, while good, do have horizon light domes, that adds light gradients to my stacks that I try to hit twice, first individually and second after RGB combination.  Note that one of my reasons for doing this, though, is that I only use synthetic luminance frames, built from RGB stacks (if you are interested in why I went that way after years of LRGB imaging, drop me a follow-up note and I will point you to some discussions here in the forum).  Anyway, I like to clean up the individual stacks before building the synlum, then hit it again after, much like I do with the RGB stack.

One thing I have been playing with is the suggestions by the Pixinsight Resources group who in a recent video suggested peppering the entire image with samples for DBE.  Very different from what I understood the model to be but interesting to test.  Using some old data to see if it offers a real improvement.

And, for me at least, my processes are built on what I learn here in the forum, followed up with trial and error (lots and lots of error  ).  I'm a liberal arts guys so I look to others for the mathematics behind it all.  As to whether you could just use PixelMath, my bet is absolutely . . . if you know how.  Again, for me at least, that's the rub.

Best,

Jim

[JoLo]

I sometimes find significant variance in the gradients between the RGB channels, especially since I generally shoot R first when the object is closer to light pollution from my front yard.  I use Jim's method of synthetic Lum, so i create it first then apply DBE to L, R, G and B.  Sometimes the gradients are similar and the same settings and placement are used, other times i have to move the samples around quite a bit between channels to get similar models.

I evaluate the results after combine, sometimes a second dose is needed, sometimes not.

I find that BN works well in most instances and is required even after DBE.  If i have good samples, ColorCalibration has little to correct.

Joe