Wipe causing AutoDev problem with ROI

[curalan]

Hi,
I've always found the basic wipe setting fine for correcting any gradients, but since getting an ASI6200 I've often had Autodev problems with a ROI. The fully calibrated image had a slight gradient top to bottom which Wipe seemed to improve. Then using AutoDev, the initial display without a ROI looks reasonable too as shown here

Autodev.jpg (435.74 KiB) Viewed 4808 times

However adding a ROI around one galaxy creates this

Autodev ROI sel 1.jpg (327.72 KiB) Viewed 4808 times

while another galaxy ROI creates this

Autodev ROI sel 2.jpg (364.56 KiB) Viewed 4808 times

If I don't use the Wipe module previously the AutoDev ROI acts normally though the slight gradient is still present as expected. I tried using different settings in Wipe including the Dark Anomoly filter but they just changed the position and shape of the dark banding. Gradient aggressiveness less than 75% default made the banding worse while 100% created some blotches on the image.
Binning the image 50% before Wipe gives more bands across the image, 4 or more distinct bands compared to the 2 shown above.
Some ASI6200 images, even with a lot of nebula, work fine with just a Basic Wipe, but others which you would think would be easier to correct with lots of background to sample, give problems. I've tried severe edge cropping but it makes no difference.
Can anybody help with explaining what's happening and how to avoid it please:think:

[admin]

Hi,

Would you be able to upload the stack so we can have a look?
What application did you use to stack these?

[curalan]

Hi Ivo,
Thanks for the quick reply. The file is too large to upload to the forum so I've uploaded it to Mediafire

https://www.mediafire.com/file/4epklmbn ... /L+180.tif

This is the L stack of a LRGB image. It consists of 180 lights stacked using AstroArt with a dark, flat and bias for flat dark.

[Burly]

What’s does your imaging train consist of, scope , ff , filters etc may be down to some sort of internal reflection.

[curalan]

Hi Burly,

It's a WO FLT-98, WO Flat 68III, Gerd Neumann M48 tilt adjuster, Zwo 7 x 2" Filter wheel and Astronomik L-3, and Deep Sky RGB Filters, ASI 6200MM Pro Full Frame Camera.
If I don't use the Wipe module beforehand these are the results of Autodev with ROI as before. They show the vertical gradient present but no odd banding, so I assumed it wasn't an optical effect.

No Wipe Autodev 1.jpg (292.59 KiB) Viewed 4756 times

No Wipe Autodev 2.jpg (352.61 KiB) Viewed 4755 times

I've tried it on different spec PCs with and without GPU acceleration but it produces the same effects. I thought perhaps the large image file size was perhaps responsible, but binning the images before Wipe just changes the banding locations and frequency.

[curalan]

If it's any help Ivo an image of the Coma Cluster gave similar banding on Wipe while an image of M42 gave no problems after Wipe. The problem images have predominantly constant black background while M42 background is very variable due to dust present.

Coma Cluster AutoDev with ROI after Wipe. The band through the middle is visible on Wipe preview though not the top right one. Heavy crop applied to avoid edge artifacts.

Coma Cluster Wipe Bad.jpg (339.39 KiB) Viewed 4705 times

M42 Autodev with ROI after Wipe. Several ROIs checked and all behaved as expected

M42 Wipe OK.jpg (194.6 KiB) Viewed 4705 times

[admin]

Many thanks for uploading this dataset. The issue appears to be that the signal is squished into only a very small part of the 16-bit dynamic range when it was saved. Can you try saving in a 32-bit FITS format (preferably integer) and see if the problem persists?

Indeed, binning to 50% (and thus increasing dynamic range 4x) seems to alleviate the issue, point to a dynamic range encoding problem.

[curalan]

Thanks for your analysis Ivo. As there is a vertical gradient on the images and the banding tends to follow a horizontal direction they could be related. AstroArt, it seems, doesn't give an option on the bit depth to save the image, it chooses a suitable bit depth to encompass all the data, and it's always been 16 bit for me. I assume the calibration is performed in floating point and then converted to integer at the end, so there must be value rounding if choosing 16 bit. I'll try another program to stack that allows 32 bit saves and report back on the result.

[curalan]

Actually the saved fits stack in AstroArt is 32 bit with this image, while the saved tif stack I posted previously is 16 bit. With my other cameras I believe it was 16 bit for both though they didn't cause problems with Startools.
However, using the 32 bit fits file, the banding is still present though not so pronounced, and it seems at twice the frequency of the 16 bit image.
Again, if I don't use Wipe beforehand, the AutoDev with ROI doesn't cause banding.
Here's a link to download the 233MB 32 bit fits file https://www.mediafire.com/file/bbinjgb5 ... 0.fit/file

Here are its results of the Autodev with ROI

32 bit ROI 1.jpg (305.28 KiB) Viewed 4603 times

32 bit ROI 2.jpg (338.78 KiB) Viewed 4603 times

Hope this helps you find the reason Ivo.

[admin]

Hope this helps you find the reason Ivo.

Thank you for uploading that. The reason remains the same; something odd is going on in the lower dynamic range and it is bit-depth related. The bands show up in the histogram as well;

Selection_579.jpg (8.38 KiB) Viewed 4595 times

This is a histogram stretched that was stretched in FilmDev without applying Wipe yet. The peaks are the bands; there are about 6 or 7 artificial transitions in the backround of various amplitudes (4 or 5 should be readily visible). It is a testament to Wipe and AutoDev's robustness and extreme efficiency in dynamic range allocation that is digs out this faint "detail" and shows it.

In terms of the histogram, Wipe merely tries to bring these peaks closer together, so that they can be shown better and needing less dynamic range allocation to do so.

As for solutions, you could try stacking with a different outlier rejection algorithm, or with Averaging. You should also make sure you image at unity gain with your camera whenever possible in order to avoid artificial truncation of bit-depth.

One other thing that is a bit puzzling about this dataset, is the stellar profiles. None of the stars over expose. It is almost as if multiple exposure times were mixed or as if a very low gain was used?