APP - Mosaic issues, 2/4 panels are green?

[ONIKKINEN]

As in the example below:

No matter what i have tried, this is what Astro pixel processor turns a mosaic into. Any ideas why? For this mosaic i did not use the local normalization correction as it creates more issues. MBB set to 15% as is roughly the overlap in the panels, this is the best integration out of many tries.

From a seams perspective, the mosaic looks good to me. Definitely acceptable, considering it was taken with a newtonian and the transparency did change a tiny bit during capture. Problem is panels 1 and 4 turn out green, while the 2 other panels look good to me.

I did manage to take 2 full runs of 30 minutes per panel on the same night, so there shouldn't be an issue with just panels 1 and 4 and all the 4 panels went through the changes during the night. I am guessing it has something to do with the composure itself, as 2 panels are almost entirely background while 2 panels are almost entirely signal. I have also posted the issue in the APP forums since someone there is most likely to shed some light on it but plenty of folks here use APP too, so might as well ask what might be the cure?

As a hack i can SCNR green to nuke the 2 bad panels, colourcalibrate, and SCNR green again and it solves the issue most of the way but i would rather solve the actual issue rather than the symptoms.

[tomato]

Use of LNC normally helps a lot with gradients when combining panels in APP, so you are at a disadvantage if this is introducing problems by using it. If you are using SCNR can I assume you have PixInsight? If so, I would recommend you try the Photometric Mosaic script. I did a 12 panel M31 and try as I might, I could not eliminate the gradients when using APP, but got a much better result from the script.

[Adreneline]

I've never done an osc mosaic in APP but I have done lots of mono mosaics.

I have found using the 'remove light pollution' tool on each pane to be very helpful and then cropping edge effects in PI before going back into APP to assemble the mosaic.

I guess you could use ABE/DBE but I find LPC works really well providing you take care to position the five (or more) boxes carefully - pretty much the same with DBE sample points I suppose.

 

[ollypenrice]

Do the two  affected panels show any sign of their green corners if stretched individually?

Olly

Edited October 26, 2022 by ollypenrice
False click

[ONIKKINEN]

4 hours ago, tomato said:

Use of LNC normally helps a lot with gradients when combining panels in APP, so you are at a disadvantage if this is introducing problems by using it. If you are using SCNR can I assume you have PixInsight? If so, I would recommend you try the Photometric Mosaic script. I did a 12 panel M31 and try as I might, I could not eliminate the gradients when using APP, but got a much better result from the script.

Apologies, should have mentioned that i used the "remove green noise" tool in Siril which is in fact just SCNR for the green channel, so no do not have PI at the moment.

1 hour ago, Adreneline said:

I've never done an osc mosaic in APP but I have done lots of mono mosaics.

I have found using the 'remove light pollution' tool on each pane to be very helpful and then cropping edge effects in PI before going back into APP to assemble the mosaic.

I guess you could use ABE/DBE but I find LPC works really well providing you take care to position the five (or more) boxes carefully - pretty much the same with DBE sample points I suppose.

I did use the "remove light pollution" tool on each of the panels before mosaic integrating them, that is what ultimately got rid of the seams and turned into the otherwise good mosaic. No luck without using this tool, visible seams on all the panels, particularly the ones that are now green.

50 minutes ago, ollypenrice said:

Do the two  affected panels show any sign of their green corners if stretched individually?

Olly

If i recomposite panels 1 and 4 back into RGB, yes they are overly green. This is something APP does behind the curtains when stacking so not sure which box to tick or untick to make it do some internal colour calibration in the stacking phase. Also should note that i did try a background extraction in Siril and then tried mosaic integrating these. But these, APP will not work with and fails at the normalization stage because of 0 measured noise and 0 signal to noise ratio or some weird error like that and the end stack is just a gray panel with every pixel at 0.025. Left that bit out as i dont understand the issue, but that might be a file format thing between different .fits files or something.

Apologies all, also left out an important detail of the workflow: All the data was indeed captured with an OSC camera, but the first step after calibration was to perform a bayer split, splitting all the subs into 4 mono ones. From APPs perspective all the work is being done on already calibrated mono files. In the mosaicing phase all the work is done to already stacked R-G-B panels. Im going to try and have APP do the work on normal OSC data now, will see if there is a change as i still think its some kind of normalization thing where APP has done some colour transformation thing wrong for the green data of the 2 affected panels. If there was no colour calibration being done on APPs behalf, all of the panels would be green as that is what OSC data should turn out as.

[tomato]

After LP and background calibration on each panel, you could try the calibrate star colours tool, to see if this evens out the green caste.

I presume you have the neutralise background button ticked on the image processing panel? This usually removes any green caste from a OSC image, and will retain this if you save your image as a stretched file.

[ONIKKINEN]

1 hour ago, tomato said:

After LP and background calibration on each panel, you could try the calibrate star colours tool, to see if this evens out the green caste.

I presume you have the neutralise background button ticked on the image processing panel? This usually removes any green caste from a OSC image, and will retain this if you save your image as a stretched file.

Neutralize background was ticked, i think. Now i have doubts so will check again. The background calibration tool can only be done for RGB images, which i couldnt do as i fed APP only mono images and the final step is mosaic integrating 12 mono stacks ( 4 panels R G B each).

Edited October 26, 2022 by ONIKKINEN

[tomato]

If the data was collected with a OSC camera, you could try stitching the RGB panels together. I have done this successfully in APP.
From memory, the Photometric Mosaic script in PI handles L and colour images rather than individual RGB channels.

[ONIKKINEN]

No cigar on stitching with OSC data not split to mono, still green in the mostly background panels, but...

Curiously i found out that it is dependent on the reference frame. Setting the reference frame to one of the background panels turns the signal panels purple, and using one of the signal panels as reference makes the background panels green. Below a quick stitch with the reference set to the bottom left panel:

It is some kind of normalization issue it seems. If i scratch my head much more i will lose the rest of what little hair remains on my head 🤪.

 

[ONIKKINEN]

And we have success! (still have some hair left, i think).

I ended up having to stack the 4 panels with no normalization, as with there will be either purple or green corners depending on the chosen reference frame. This created a flawed mosaic with visible seams around the darker edges but the core was equalized in colour and brightness, i LP removed this as much as possible in APP to try to even out the edges, but no working mosaic yet. I then cropped the flawed mosaic so that only the core and a little bit of background on both sides remained and used this new panel as a center reference frame. This was integrateable without normalization, but with LNC 1 set to 3 iterations. This leaves me with an image where all of M31 is equal in levels and colour and no obvious seams without a nuclear stretch. This new integration was salvageable with Siril background extraction so that all of the image looks more or less balanced. Crude autostretch below:

Now, its not a finished image yet of course but the mosaicing did end up working somehow. Now that i know it can be worked with i will attempt to get another 4 hours if a decent moonless night arises before M31 sets as the year progresses, and more importantly, set aside some time to shoot an actual center panel too to try and figure out if that was the issue in the end.