How best combine two stacks

[GaryP]

I have two stacks of the same subject that I would like to integrate because PI would not integrate all 72 subs in one batch on my system, as explained in a recent post. They need to be cropped a bit. My questions are:

1. Are the stacks already registered by virtue of the fact that the subs were registered?
2. Does it matter if they are cropped before or after after integration?
3. If they are preregistered and then cropped, is it necessary to reregister them before integration?

[bitli]

• If your registered both stacks with the same reference image, they are already registered together.  If you registered them independently, then they are not.  I recommend that you integrate both stacks on the same image, StarAlignement can align against any image, even if the image is not part of the stack you register (align).
• Not much.  Since the last version the statistics can ignore the cropped (dark) area.  The noise evaluation is also not very sensitive to it. The location/dispersion statistics may give slightly different results, but I do not think it matters in real case. You can always crop first if you want, it is not very long.
• No, not if the images were aligned towards the same reference and cropped with the same instance of dynamic crop.  It is easy to check, just put both image on top of each other and blink them with Ctrl/Next.

Note that with two images you can just do an average of them (you need at least 3 to do rejection). But the rejection should already be quite good within the two stacks.
-- bitli

[Juan Conejero]

All noise evaluation and statistics calculation routines reject pixels outside the [0.00002,0.99998] range. So extremely dark and bright pixels are ignored and hence have no effect. On the other hand, extensive use of robust statistics ensures that outliers have no influence on the computed estimates of location and scale (dispersion).

[Carlos Milovic]

Could you add a parameter somewhere to disable this?

[GaryP]

Thanks for the useful answers. I don't understand CM's in-joke, but I suppose you see these questions asked frequently. I did do a search, but most of the answers seemed aimed at problems integrating monochrome frames.

[bitli]

I guess the joke was more related to me not being sure that the evaluation is insensitive to the order of the crop, and Juan stating (as he did a few time) that the evaluations are not sensitive due to the limits 00.-0.98 (they are hardcoded in a few places).  Maybe PI is too good and we apply folkloric tricks (like truncating before integration) that are long needed ...
The short answer is: truncate at the end. If you create a single integrated image (DSLR), you can use the dark rejection mask to find the area to truncate very easily.
The second point is: do not hesitate to ask questions if you do not find the answer already.  And sometime even if you do, you may get other opinions, alternate ways doing things, and so on.... PI is a living object.
-- bitli