optimising the dull job of image stacking

[stephen63]

OK, Chris
I guess read noise IS the limiting factor. So, after the images are stacked, as long as the SNR is at least 3 then it can be separated from the noise.

[Chris Peterson]

OK, Chris
I guess read noise IS the limiting factor. So, after the images are stacked, as long as the SNR is at least 3 then it can be separated from the noise.

S/N > 3 is commonly used as a target in scientific imaging. For aesthetic imaging, people usually want much larger values.

[Nitpicker]

S/N > 3 is commonly used as a target in scientific imaging. For aesthetic imaging, people usually want much larger values.

No wonder I still think my humble photos still have some slight scientific merit. I always try to remind myself that the universe is a noisy place.

[Beyond]

S/N > 3 is commonly used as a target in scientific imaging. For aesthetic imaging, people usually want much larger values.

No wonder I still think my humble photos still have some slight scientific merit. I always try to remind myself that the universe is a noisy place.

Yeah It's a good thing 'space' is a hard vacuum, or we wouldn't be able to sleep at night.

[geckzilla]

Coincidentally, I am messing around with four pretty bad exposures from the HLA with low S/N ratio and a crapload of cosmic rays. I'm getting some interesting results by taking four different possible combinations of pairs of them and using minimum which I presume is some kind of floor function and then creating a median stack out of those four combinations. I thought it might be a good way to recover some of the signal loss from using only a minimum stack.

What software are you using? Because median processing increases noise, the method has largely been replaced with more complex stacking algorithms, which do a better job of eliminating single frame artifacts while preserving the overall S/N similar to summing. Sigma mask, sigma clip, and similar algorithms are generally better choices.

I have been thinking about this thread as well as this post since yesterday. I downloaded CCDStack since it offers these more complex algorithms and tried them out. After a crash course with their tutorial, I recreated my homebaked Photoshop method along with two sum stacks using STD sigma data rejection and Poisson sigma data rejection. All results looked "good" as far as this particular data goes (it's the faint, outer shells of the Calabash Nebula. It's hopelessly noisy and chock full of CRs.) Interestingly to me, the Poisson sigma data rejection looked very similar to my home baked idea of pairs of minimum stacks combined into a median stack.

Visually, it's hard for me to determine the S/N so I used the software calculation. Poisson sigma sum came in last at 0.31, my silly median minimum second at 0.34, and STD sigma sum on top at 0.39 for the same rectangle around the object for each.

The bad part about the STD sigma sum stack is that it also left quite a few pieces of cosmic rays for whatever reason, so even though the data overall is less noisy, for aesthetic purposes it might be harder to work with. Then again, maybe a day worth of learning and tinkering with it wasn't enough for me to know how to fine tune it.

PS - I hope you don't mind me interjecting this in your thread, Nitpicker.

[Chris Peterson]

The bad part about the STD sigma sum stack is that it also left quite a few pieces of cosmic rays for whatever reason, so even though the data overall is less noisy, for aesthetic purposes it might be harder to work with. Then again, maybe a day worth of learning and tinkering with it wasn't enough for me to know how to fine tune it.

When it comes to rejecting transient artifacts, it helps to have more frames. You don't usually have that luxury with HST data.

[geckzilla]

The bad part about the STD sigma sum stack is that it also left quite a few pieces of cosmic rays for whatever reason, so even though the data overall is less noisy, for aesthetic purposes it might be harder to work with. Then again, maybe a day worth of learning and tinkering with it wasn't enough for me to know how to fine tune it.

When it comes to rejecting transient artifacts, it helps to have more frames. You don't usually have that luxury with HST data.

This one actually had 7 exposures which is a little more than usual. One of them was so short it seemed to degrade the whole stack even with sum so I left it out. It needs a couple more hour long exposures and some good luck with the CRs smooth it out. No wonder they had to point the thing so long for the ultra deep fields, which it seems like they are going to do again. Going to watch the Hubble Hangout in three hours to find out more.