Stacking Test

Tony,

The main thing that counts is SNR measured in the dark regions and I can’t get at that number from what you posted. The stacking parameters allow for different means of weighting the average and the results may vary significantly depending on how you have that set up. The results will also vary somewhat depending on how you have the rejection filters configured—particularly at the low end. PI also allows for different ways to normalize frames within the stack to statistically “align” the data through the stack. How did you handle that normalization? There may be some slightly different ways to implement a weighted average but if you turn off all the filters, do a simple unweighted average, and disable any renormalization, you should get nearly identical results. Otherwise, I’d be surprised if you can ever get exactly the same results out of these three programs. The biggest concern should be showing how to properly configure the stacking parameters to maximize SNR—again as measured in the dark regions.

John

Tony Gondola · Jun 29, 2026 at 04:11 PM

John Hayes · Jun 29, 2026, 03:35 PM

Tony,

The main thing that counts is SNR measured in the dark regions and I can’t get at that number from what you posted. The stacking parameters allow for different means of weighting the average and the results may vary significantly depending on how you have that set up. The results will also vary somewhat depending on how you have the rejection filters configured—particularly at the low end. PI also allows for different ways to normalize frames within the stack to statistically “align” the data through the stack. How did you handle that normalization? There may be some slightly different ways to implement a weighted average but if you turn off all the filters, do a simple unweighted average, and disable any renormalization, you should get nearly identical results. Otherwise, I’d be surprised if you can ever get exactly the same results out of these three programs. The biggest concern should be showing how to properly configure the stacking parameters to maximize SNR—again as measured in the dark regions.

John

That’s very helpful John as the reason I posted this was to find a meaningful way to understand the results. What I was going for was to see if there are any differences when using the default setups for the various programs simply because it’s a reflection of how they will be used by most people. Since the ability to customize the process varies a lot between the programs, I thought that would be the best, though not scientific, approach. I agree about SNR and will take a look at what options have commonality between the three programs and what the tradeoffs might be. I would have expected that the default settings would provide the most balanced result but that might not be a good assumption.

Unfortunately, PI doesn’t really have what I would call “default settings”. They sometimes will set a parameter and indicate that it should work in most cases but they leave it up to the user to understand what the parameters mean. Things like the normalization and weighting settings can make a significant difference in what you get and that’s one of the challenges for most folks when learning PI. You have to understand the entire data flow to get things properly optimized. It’s also important to understand that just because they provide tools to measure data statistics, you can’t just apply them to an entire image and get a meaningful result. That’s why I’ve pointed out that in order to correctly analyze the SNR (for example), you need to analyze a statistically significant small dark patch that doesn’t contain very many stars. That makes the kind of comparison that you are trying to do a bit tricky…and time consuming.

John

Tony Gondola · Jun 30, 2026 at 12:27 AM

Ok, here’s the calculated S/N for each stacked result:

Siril - 15.3

SASpro - 15.0

PI - 17.5

That’s interesting. So, how did you end up setting the parameters and measuring SNR for the three results?

John

Tony Gondola · Jun 30, 2026 at 03:52 PM

I took the mean for a faint other portion of the Galaxy and for the sky background, subtracted the background and divided by the standard deviation.

Thanks for that explanation Tony. As long as you used the identical regions in each image, that should work well.

John