15th Feb 2024: Astro Pixel Processor 2.0.0-beta29 released - macOS native File Chooser, macOS CMD-Q fixed, read-only Fits on network fixed and other bug fixes
7th December 2023: added payment option Alipay to purchase Astro Pixel Processor from China, Hong Kong, Macau, Taiwan, Korea, Japan and other countries where Alipay is used.
Whenever I stack subs from multiple nights, if I forget even a single frame with parts of the foreground (e.g. roofs of distant houses, trees, entering into the frame), that generates a very strong dark (or bright) blurry spot that ruins the entire integration. Playing with rejection parameters does not seem to have an impact (at least to the extent of the simple tests that I've run).
I then simply remove that sub and just integrate the rest without issues.
My problem is that right now a construction site has popped up and I have relatively frequently small parts of the image that are covered by a crane passing by an edge of the image. It is not covering the objects I'm interested in, but it is present in enough subs that i lose a significant amount of integration time if I just remove them. (1-2 hours on a typical project).
Is there a way to mask out the offending parts so that they are simply ignored by APP when stacking? I imagine that just putting all pixels to 0 might do the trick? (it is, after all, what happens when integrating mosaic panels that are registered together, the "rest of the image" is simply black. Is there something more that should be done? (alpha channel? some masking bits that might not be obvious?). What would be a good way of creating such a mask on ~50 subs? (not an issue if I need to do it 50 times, it's more a question of "which software would be most appropriate"?)
Thank you in advance!
After a bit of testing:
Set all the pixels "to be masked" to 0 in photoshop, saved and reloaded in APP. I have 113 files in total, with 41 masked out.
-
- With automatic integration the averaging is set automatically to average, which leads to very strong artefacts due to the fact that more than 1/3 of images have blanked out pixels.
- Setting integration to median improves things dramatically, nevertheless, the large portion of images at zero in the masked out region means that the median values being selected are on the lower end of the diistribution of "real pixels".
- Trying with Maximum yields a perfect image in the masked out regions (it is ignoring the blacked out images) but the rest of the image is a mess
- Playing around with pixel rejection, I lowered dramatically the low kappa threshold, to only select a few pixels around the "real pixels" values, but there was no real difference against the standard median
And Local Normalisation saves the day! (or rather the lack of!)
I suspected that local normalisation would mean that the region with my masking would be analysed on its own, meaning that the gap between masked pixels and real data would be smaller than if all pixels were analysed together. And indeed that works out pretty well. The darker areas disappear completely and I've been able to add ~1.3h of data to my "clean" 2.4h. Given the challenges of getting data (and good weather!) this is a massive improvement!
Now I regret deleting all my previous "crane and other stuff" images!
And a bit more testing shows that swapping Median for Average (but still not using Local Normalisation Rejection) basically gives exactly the same results!
Yes, the algorithms are not able to recognize objects in the frame, this is probably not possible as that will be extremely variable. Thanks for the elaborate analysis! This is really interesting, local normalization indeed tries to get regions "averaged" in background illumination which usually works very well in true astro fields of view. 😉
