Large X shaped pattern tiled across images

[mburge]

Any idea where this large X shaped pattern tiled across the images is originating? I am stacking 5 images (taking the maximum) with registration and my guess is it is originating in that process. The raw images are being captured in 16 bit from an ASI294 with 60 second exposures. I'll try to chase it down tomorrow but thought you might have an idea off hand. I'm saving images as jpg's using the default settings. I attached a sample in PNG format.

[aaronwmorris]

Are you using the CLAHE for contrast enhancement?

[mburge]

No, it is not enabled, in addition to the stack with registration, the only things on are AWB, saturation at 1.5, average neutral SCNR, and stretching. All at defaults. I defocused the camera to check if it was in the light path, but no luck, here's an externally contrast enhanced detail:

[mburge]

Looks like it is an artifact of the registration process. I disabled all stacking. No artifacts. Enabled stacking 5 with maximum but w/out registration. Got images with star trails consisting of 5 stars, as you'd expect with maximum and no registration. Enabled registration and artifacts returned. My guess is that when solving for the transformation matrix it is iteratively alternating between close solutions, maybe just because of precision. Kind of neat actually, that is why the structure is so regular. I'll look tomorrow. By the way, love your system.

[aaronwmorris]

Thank you!

I could barely see the pattern on my laptop last night (but I could see it), but I can see it much better on my workstation.

I have used registration quite a few times, although, I rarely leave it on. It has to be a clear night. However, I have never seen that. My cameras are not as high resolution as the ASI294.

https://github.com/quatrope/astroalign is the module used for registration. I will be the first to admit that the registration can be less than perfect. All sky images have enough distortion and enough variations due to planes and satellites that misalignments are very common. After I implemented the filter to check for rotations exceeding a standard deviation, misalignments are less common now.

I wonder if this pattern is some interaction due to the quad bayer matrix on the IMX294.

[mburge]

I had a chance to run some tests and it is definitely a result of registration.

Split screen with registration

Split screen with just stacking

I haven't found where it is yet, I think you're right, it's an artifact of the quad mask / 16 bit format. Solution for now is to just turn off stacking and registration.