[codemonkey (Lee)]

We've had some pretty average conditions here lately, even though the skies have been clear. Bad seeing some nights, very windy others, sometimes both simultaneously.

I'd captured a few good subs of my next target, but many more that were less than great and I wondered if I could use these less than great subs to reduce noise, whilst not having a negative impact on star sizes and finer detail.

I always weight my images prior to integrating them such that images with a lower FWHM and eccentricity have more weight than subs with higher values, and I still applied this to the examples attached.

This time I split up the subs. Anything less than or equal to 2.4" FWHM went into the good pile, everything else went into the bad pile. The bad pile contained subs up to 3.9". I then integrated the bad pile and good pile separately.

After integrating, I took the bad integration, cloned it and did a histogram stretch on it such that the background was very dark, but stars and the subject were well stretched.

I also cloned the good integration, then masked the image with the stretched bad integration (inverted).

From here, I used PixelMath to add the bad integration and the good integration together, weighted more strongly towards the bad one (which had many more subs, and thus less noise), and applied this to the masked good integration.

The result? The noise went down significantly, but I kept the nice stars and (most of) the details from the good integration. I believe I managed to retain 90% of the SNR when compared to an integration with the entire set of subs.

FWHM of the "all subs" integration was 2.46176", "good only" was 2.32288" and the merged one was 2.18288". That's not a typo, according to PI, the merged one had better FWHM. Not sure what to make of that.

Anyway, you could argue that it's not worth it to lose that bit of SNR, or that the marginal decrease in FWHM is simply not worth the effort, but I thought it was an interesting little experiment.

Of course, you could do many things differently. The masking in this case was entirely based on the intensity, but that may not always be ideal, you might have lots of dim, fine detail. You'd probably also be better integrating all the subs and using that instead of a manually weighted sum of the good/bad stacks for the lower signal areas.

Attached is a copy of one integration with all subs, another using the merged approach described here, and then a stretched copy of the difference which highlights the impact on star sizes.