Multiple copies of the same exposure, but randomizing the noise?

Question

As a budding astrophotographer, I find my results are still largely hit or miss. Sometimes the standard procedure of stacking multiple exposures to randomize the noise and boost the signal works great, and other times the stars themselves get washed out by the process. Stacking the layers accurately can be next to impossible when shooting at high or low celestial latitudes because the image has to be rotated as well as laterally shifted. Moreover, the center of rotation (a celestial pole) is usually far outside the actual image.

But here's an idea: What we could just duplicate a single exposure, but somehow randomize the noise from one copy to the next? That would eliminate the need to stack different exposures.

Is there any software that can do this?

Answer 1

To achieve what you're thinking of you would have to know what the noise was.

If you knew what the noise was then you could just remove that to get clean images.

Answer 2

Image stacking works to reduce noise because the noise is random — or at least, ideally so — while the stars are (famously) constant.

That means that (once you've corrected for rotation) the stars will be in every photo. But noise — at least, the kinds of noise that this can correct for — is already random fluctuations... maybe there in one image, and not in the next. So, when you add images together, the signal increases with each image, and the noise only with some of them.

If you just have one image, you don't have this way of distinguishing. Image stacking only works because you have multiple separate exposures.

Answer 3

What you're asking about is generally called star-eating. From an algorithm point of view, in an astrophotograph, how can software determine what is signal (that is, stars) vs. what is noise?

A simple method would be to just call anything below a certain intensity level as noise, and cut it all off below that level. In essence, that's the same as throwing away dynamic range. That is, for a 12 bit conversion, its throwing away the least significant n bits.

Aside from throwing away dynamic range, it also has the side effect of throwing away actual signal living down amongst the noise. What you're left with is an image that has far fewer stars than you expect to see — the stars were eaten by the algorithm.

Several cameras do this internally on die, before analog to digital conversion, in an attempt to improve their low-light noise performance. These cameras have reputations as star eaters.

Astrophotographers employ lots of tricks to improve their output. In addition to image stacking, you might consider obtaining or building a tracking equatorial mount aligns the camera's mount with the earth's rotation axis. This reduces the difficulty of tracking the sky down to controlling a single rotation axis at a constant speed. This is why most telescopes use an equatorial or German-equatorial mount.

For more on improving you astrophotography in general, see also:

• Why am I unable to capture as many stars as I can see with my eyes with my new DSLR?
• What should I look for in a camera for shooting in bulb mode for astrophotography?
• Astrophotography: Balancing the ISO value
• How can I achieve more clarity in my photos of the moon?

Regarding tracking mounts, see:

• How do star trackers work to take a photo for 5 minutes without blur, and are they worth it?
• How can I avoid star trails without an expensive tracking mount?
• What equipment do I need for astrophotography?
• What features to look for in a motorized DSLR mount for astrophotography?
• Which tracking motor to use for low-end wide-angle astrophotography?

Answer 4

As others have pointed out, what you describe is not possible. If you could randomize the noise, it would mean that you can already detect what is noise and what is not. Then you could just as easily remove the noise and keep just the signal. But you do not know what the noise is and what is actually signal, that's why people stack photos - that actually "randomizes" the noise, because noise is random, so if you keep adding noise+signal during the stacking, the noise will cancel itself out because it is random, while the non-random signal will just keep adding up, thus increasing the signal-to-noise ratio, leading to a nicer image.

What seems to be happening is that you have problems with image alignment.

You say that "stacking the layers accurately can be next to impossible", but this is most definitely not true. Good astrophotography stacking software should help you with that. Maybe you're trying to do it in a program which is not very suitable. Or maybe you're already using DeepSkyStacker (pretty much the de facto standard SW for stacking), but doing something wrong.

If you're unable to figure out what the problem is on this site, maybe you should visit an astrophotography forum, explain your situation and provide the source files, and somebody there will be able to tell you why it does not work.