Studies of an extreme case in mosaic processing, where all standard methods fail

Curiously, I did something on the same lines (minus the background matching PI script) so I kind was expecting you moved on similar lines. Thanks for the write-up and well done!

donld · Jul 13, 2026 at 12:44 PM

Very interesting, thanks. I’m curious how a Photoshop workflow would proceed after the stretch.

Hi,

I do most of the retouching in Photoshop. I do curve and level stretching, and saturation boost. I always mix a global saturation boost and a luminosity-weighted saturate boos (brighter parts get more saturation boost while faintest parts get none).

I apply unsharp masking to sharpen details (in addition to BXT in PI), and this is luminosity weighted as well, to avoid amplifying noise. Then I add multiple layers of high-pass filtering (with different filtering radius) to amplify the larger-scale nebula structures. If just NXT is not enough, in Photoshop, I do my own noise reduction using series of surface blur in Lab space and apply to original image with again a luminosity weighting (more noise reduction in the fainter part). The sharpening, high-pass filtering, and noise reduction are all done on starless images to avoid artifacts around stars or blurring the stars. Some image defects are also best corrected on the starless version.

Final retouching may include further star shrinking using a combination of minimum filter (similar to morphological transform in PI) and small-radius unsharp masking, with a mask to act on stars only. And then color-balancing and further curves/levels/saturation to adjust the look and feel of the image.

donld · Jul 13, 2026 at 12:44 PM

Non-astro mosaics typically are shot with more overlap than astro. It would be interesting to learn whether more frames would make the problems you encountered better or worse.

A lot of mosaic defects can be remedied with large overlapping between panels or more panels. Unfortunately this does not apply here. The example here contains a 4×4 mosaic, and then a 5×5 mosaic offset by 0.5x the FoV of an individual panel in both E-W and N-S directions. So every point on the sky is covered by at least 2 different panels. And the mosaicking using standard algorithms still shows the problems.

Moreover, what I didn’t say is, the imaging was actually done by 3 scopes/lens simultaneously. All of them have 250mm focal length. One is with a median format camera (44mm*33mm sensor), and two are with FF cameras (24mm*36mm sensors). The two FF combined give an FoV comparable to the median format one. So actually, after including all data, any point on the sky is covered by at least four different panels. And this doesn’t lead to any better results.

The problem I described here essentially has nothing to do with how the panels overlap with each other and how many panels are there.

donld · Jul 13, 2026 at 03:29 PM

Moving towards first principals, the data between panels doesn’t match well when merged. So either the capture is different or the stretch before merge is different. That difference isn’t noticeable in low contrast, but must be present.

Oh well… I think you have a lot to read about basic astronomical mosaic workflows.

Dear Wei-Hao,

What a fascinating and informative read. I have been working with mosaics for a few years now. The biggest challenged faced has been misaiming the 1200 or so OSC fields for the full astrobin survey that @James Tickner is working on.

I have had some success at the ~25 field level with the following workflow

PI. GradientCorrection: SpecPhotFluxCal: SpecPhotColCal: BXT (Correct only)

APP. Mosaic Mode; with LNC 2nd order/MB Blending at 15%

The mosaic can be now be stretched in PI (using GHS or MAS) without showing seams. I have a few examples on my posts.

APP fails at larger sizes - but primarily with the astrometry over such a wide field. I think @James Tickner is closing in on his own solution for a larger number of fields and you should compare notes.

I had given up on GHS in PI, but after reading your paper, I may reinvestigate. I also tried using MARS with my 40mm data, but found it didn’t help. Like you, I fought it was the stars that were the issue. How? Not sure? But they do vary in structure from field to field - particularly if neighbouring fields are taken over a variety of nights.

Brian

Very nicely done Wei-Hao! I enjoyed your write up and I look forward to seeing the final result. With all of the multi-aperture systems coming online, I would encourage you to consider publishing a paper about this subject. Your work certainly would have been very worthy of a poster paper at the Copenhagen meeting.

John

Wei-Hao Wang · Jul 14, 2026 at 04:08 PM

John Hayes · Jul 14, 2026 at 02:54 PM

Very nicely done Wei-Hao! I enjoyed your write up and I look forward to seeing the final result. With all of the multi-aperture systems coming online, I would encourage you to consider publishing a paper about this subject. Your work certainly would have been very worthy of a poster paper at the Copenhagen meeting.

John

Hi John,

Thanks. That’s an interesting suggestion. I suppose the Copenhagen meeting is mainly for instrumentation, i.e., hardware. There some colleagues from my institute in that meeting. They all work on instrumentation or observatory operation. I myself pretty much only attend conferences for astronomical studies. That being said, long time ago, I attended one of the Astronomical Data Analysis Software & Systems (ADASS) conferences and gave a talk there. It’s about my mosaic processing pipeline for infrared mosaic cameras on 4 to 10 meter class telescopes. But that’s for processional research. I am not sure if there is a conference that accepts papers about amateur data processing. Even if there is one, since it’s about amateur works, I can’t use my grant to attend a conference to present such a paper. So it has to be in east Asia, so I can more easily pay by myself.

Talking about multi-aperture systems, a lot of them are for astronomical research or SSA. They are perhaps less about making nice looking mosaics. But anyway, if any of them want to create big mosaics and struggle for similar issues, I hope my document helps to point to some good directions.

Hi Wei-Hao,

The emphasis at the SPIE meeting is indeed on instrumentation; however, the range in sophistication among the poster papers was quite large. Frankly, I could have submitted a paper about some of the stuff that I’m doing on my own scope and it would have been relevant, new, and useful to some of the folks there. I totally understand your point about the science images produced by multi-aperture systems vs producing a “pretty” mosaic but again, that’s what the poster papers are good for. Someone, somewhere is working on trying to figure out a solution to a problem that you’ve already solved! I completely agree that it’s not a “science level” topic, but you might be surprised by how much interest it might get—particularly if you present it with even a hint of connection to some of these new programs. I saw numerous papers where the authors churned and churned to analyze to death something very simple. I saw others about some totally trivial thing—like the OAG poster paper that I mentioned. Clearly, the best subjects for a paper are obviously about new, novel, and maybe even revolutionary ideas; but IMO, this sort of conference is also about the exchange of methods and techniques that can be helpful to others and that’s what you’ve done here. It’s just something to consider…

John

John Hayes · Jul 14, 2026 at 05:05 PM

The emphasis at the SPIE meeting is indeed on instrumentation; however, the range in sophistication among the poster papers was quite large. Frankly, I could have submitted a paper about some of the stuff that I’m doing on my own scope and it would have been relevant, new, and useful to some of the folks there. I totally understand your point about the science images produced by multi-aperture systems vs producing a “pretty” mosaic but again, that’s what the poster papers are good for. Someone, somewhere is working on trying to figure out a solution to a problem that you’ve already solved! I completely agree that it’s not a “science level” topic, but you might be surprised by how much interest it might get—particularly if you present it with even a hint of connection to some of these new programs. I saw numerous papers where the authors churned and churned to analyze to death something very simple. I saw others about some totally trivial thing—like the OAG poster paper that I mentioned. Clearly, the best subjects for a paper are obviously about new, novel, and maybe even revolutionary ideas; but IMO, this sort of conference is also about the exchange of methods and techniques that can be helpful to others and that’s what you’ve done here. It’s just something to consider…

John

Thanks. Looks like it’s a conference that I have to go at least once. I like the diverse topics that you mentioned. The science conferences that I usually attend have focused topics, and sometimes that can be boring.

Wei-Hao Wang · Jul 14, 2026 at 09:12 PM

GalacticRAVE · Jul 14, 2026 at 05:29 PM

Excellent write-up Wei-Hao, can’t wait to see the final mosaic! CS Matthias

Thank you. I am working on it. Retouching of such a big image takes lot of time. Earlier this month, I thought I was able to publish it. Then I found something in the near-final result that I was not quite satisfied. That dissatisfaction required some rework at the very early stage. So I went back and do everything all over again. I hope I don’t need to do this again.

Oh man…I really hate it when that happens!

📷 abc_cross.jpgHere is the starless version of the Southern Cross region from the ABC survey fields (taken by myself and @James Tickner ). I posted the image (with stars) back in Nov last year

📷 cross_ha_nostars.jpgAnd the Ha image mosaic of (roughly) the same area - posted a couple of months ago with the stars.

Hi Wei Hao,

No both were done using the method outlined above. Initial field processing in PI → Mosaicing in APP → stretching in PI.

I think James’ method will work even better. APP is still limited to about 25 frames, before it starts to scale down automatically and/or fail due to the astrometric solution over such a wide area.

Wei-Hao Wang · Jul 16, 2026 at 02:26 AM

Ah! I see. Then it’s interesting that APP works better for you (in terms of creating a seamless mosaic even after star removal).

I vaguely remember somewhere during the APP process, it warned me the mosaic is too big. But I didn’t check if it scaled the mosaic down. If it’s the case (scaling down), then APP isn’t the best tool for big mosaics. We probably have to work completely in PI if we want to avoid reducing the image resolution.

I suspect APP isn’t ultimately the best tool for mosaics. But it is the best available tool I have found to date.

Nevertheless, I am in awe of what APP is currently able to do with mosaicing and only wish PI was half as good. In my experience it is not. I never had much luck with GMM. PMM is good, but tedious beyond belief to do more than a few frames. Also PMM does weird thing with field boundaries when images are rotated more than a few degrees with respect to each other.

APP is also a bit of a “black box” to me, but then again most software programs are. I feel were are getting closer to the perfect mosaicing programs - which must in some way also be driven by commercial requirements for the explosion in smart telescopes.