Stacking fails with PixInsight [Deep Sky] Processing techniques · Michael E. · 6/18/2023 · 10 · 341 · 0

When stacking images taken with my two scopes, a 90/500mm Refractor and a 180/1800mm Mak, I have noticed that images from the latter often fail to stack using the Weighted Batch Processing (WBPP) in PI, whereas the images from the first one don't cause any issues. The failure usually occurs during the two local normalization steps, resulting in no master file and displaying the following error message:

!!! Error: Error applying local normalization to light frames. This group will be skipped.

I suspect that the problem might be to the larger scope being oversampled at 0.43" per pixel causing PI to struggle with the bloated stars.
However, I have never encountered this issue when using DeepSkyStacker.

==> Does anyone know if there is a way to resolve this problem by adjusting certain parameters within WBPP?

Michael:
I suspect that the problem might be to the larger scope being oversampled at 0.43" per pixel causing PI to struggle with the bloated stars.
However, I have never encountered this issue when using DeepSkyStacker.

Oversampling does not create bloated stars.  If you take your oversampled image and resample it to a less sampled image for comparison, you will find the oversampled image actually has smaller stars... seeing being the same.  The idea that oversampling produces bloated stars is one of those "old wives tales" in astro-imaging.  They just look bigger because, well.. you are more sampled.

Michael:

andrea tasselli:
Avoiding using WBPP altogether would help in zeroing the origin of the issue. Binning (also post-fact) would also help since there is now way you can achieve the underlying resolution of the sampling.

Thanks for your response!
2x2 binning sounds like a good idea and I was about to give it a try, but unfortunately couldn't find any way to produce 2x2 binned images from the original 1x1 in PI.
Maybe you could give me a hint on how you would do it?

However, I did manage to achieve it in a different way:
Within the WBPP I changed the Debayer setting from VNG (default) to SuperPixel (ref: https://pixinsight.com/doc/tools/Debayer/Debayer.html) which smoothly ran through the entire script and consequently resulting in an image with a reduced resolution (pixel wise)!

When you mentioned avoiding the WBPP Script altogether, did you mean using another tool like DSS as I did initially, or manually performing all the stacking subprocesses within PI?

Yes, I meant doing eveything step-by-step without relaying on a script. There is a way to do binning in PI even with Bayer matrix but at the moment I can't recall what it was. I'll be back once it pops back in my mind.

Jure Menart:
I used IntegerResample process in PI when I needed to do 2x2 binning in postprocessing, it's quite straightforward to use it.

This isn't the same of binning an OSC image with Bayer matrix, as you would need to interpolate first and then average 4 pixels across for each color plane.

Re: binning an OSC post-fact

I originally did it MATLAB as I had only to bin a single master flat so that won't work despite being quite nifty. Truth is there is no strightforward way to perform it in PI so it pays to bin "in camera" big time. A work-around is as follows:

1. Pre-process and do cosmetic-correction on the original files
2. Use SplitCFA process adding all the files of the sequence
3. For each channel/directory (there will be 4 of them, named [0],[1],[2],[3] for RGGB) create an image container specifying both source and destination folders
4. Create a process icon using Geometry/IntegerResample, with downsample mode:Average
5. Drop the Process Icon created in the step above onto each of the image containers icons created at step 3

Now you got  4 streams of files containing the R G(1) G(2) B colour planes "binned" 2x2 so the file size should be 1/4 of the orginal.
Keep processing in the usual way as if you had a monochrome camera with colour filters. Align using the best G(1)/G(2) channel for all of them. Normalize and Integrate each of them 4 streams and finally average (PixelMath) the G(1) and G(2) color planes together to create the "master" G frame. Finally create the RGB composite and save.

My original method developed in MATLAB (but it might as well be done in PixelMath but I don't know how) recreates a CFA array by binning a 4x4 matrix into a 2x2 matrix, as if it was done in camera, so it is actual binning. The other method described by myself above and the PI's SuperPixel (I didn't know this one, so today is school day!) should produce essentially the same result and they both do not bin per se except for the G channel which gets a boost in SNR.