Noise cancellation tools?

[Artlav]

Are there any tools - programs or photoshop plug-ins/methods - to remove noise from images by cancellation?

That is, give 2 images with fixed objects and a different sets of noise to produce a single image containing only fixed features between the two.

 

[agentgonzo]

If the images are exactly alike, then you should be able to combine them in photoshop using the average of the two images (or use registax to do this automatically). You can then probably use the noise reduction feature of photoshop on the resultant image to sharpen it up a bit again.

 

[Redburne]

I think you're looking for a process that is usually called image stacking. There are quite advanced methods for astrophotography.

If you only need to remove the noise by averaging, you can layer the images in photoshop with transparency (Bottom layer 100%, next layer 50%, 3rd layer 25%, 4th layer 12%, and so on).

If you're after the pixels that changed/did not change, playing around with the layer mode "difference" or "subtraction" (or whatever it is called) could help.

BTW, newer versions of Photoshop can try to import a number of images and align them automatically, if you need that.

 

[Artlav]

If the images are exactly alike, then you should be able to combine them in photoshop using the average of the two images (or use registax to do this automatically). You can then probably use the noise reduction feature of photoshop on the resultant image to sharpen it up a bit again.

If you only need to remove the noise by averaging, you can layer the images in photoshop with transparency (Bottom layer 100%, next layer 50%, 3rd layer 25%, 4th layer 12%, and so on).

That way helps a lot, albeit not perfect:

And some adjustments:

Still, it looks like an unsharp average rather than a picture of the sky, and it does nothing to light pollution.

I think you're looking for a process that is usually called image stacking. There are quite advanced methods for astrophotography.

Thank you, that is something googleable at last.

If you're after the pixels that changed/did not change, playing around with the layer mode "difference" or "subtraction" (or whatever it is called) could help.

Difference and exclusion. Unfortunately, they both produce a mathematical result of little use.

BTW, newer versions of Photoshop can try to import a number of images and align them automatically, if you need that.

That would have been nice, how to make it do that is CS2?

 

[Redburne]

You may be out of luck in CS2.

CS3 has the function "Files/Scripts/Load Files into Stack ..." which loads the files as layers, optionally aligning them.

Edit: There is a forum message with a script which allegedly also works in CS2, but I'm a bit sceptical - especially about the alignment part.

 

[TSPenguin]

As mentioned before, registax is made for exactly that purpose. It works best on planetary images but I had some good results on sky images too.

Regarding light polution, you have very limited options. If you have mostly the same light poluting your location you could use a laser with an interfering wavelength, but I haven't tried that.

 

[agentgonzo]

Still, it looks like an unsharp average rather than a picture of the sky, and it does nothing to light pollution.

You'll not be able to get rid of light pollution by these methods, as the light pollution is an inherent part of the scene that you took the photos of. There are ways to get rid of light pollution. It basically requires taking a copy of the image as a new layer and applying a large gaussian blur to it (that essentially blurs out all stars. Since the light polliution is diffuse anyway, the blur doesn't really affect it so you end up with an image that is pretty much just the light pollution). You can then subtract this blurred image from the original which essentially gets rid of the light pollution. Google some words for better information and, I'm sure, lots of blogs on the subject.

 

[Artlav]

As mentioned before, registax is made for exactly that purpose. It works best on planetary images but I had some good results on sky images too.

A solid one for automation, but the results are far from a manual photoshop.
Never the less, thank you for that one, could be useful.

If you have mostly the same light poluting your location you could use a laser with an interfering wavelength, but I haven't tried that.

Um. Since Mitinobazar was turned into a parking lot i don't think there is a place around here where you can buy something of that description for no money.

It basically requires taking a copy of the image as a new layer and applying a large gaussian blur to it (that essentially blurs out all stars. Since the light polliution is diffuse anyway, the blur doesn't really affect it so you end up with an image that is pretty much just the light pollution). You can then subtract this blurred image from the original which essentially gets rid of the light pollution.

Now, THAT is what i asked for.

Result:

Compared to a reference generated by Stellarium:

Thank you for the neat trick...


On a more original topic question - now i remembered where i got the idea from - i've seen an article with a detailed making process of the Hubble images, which consisted of having 2 images of the same object being taken, then combined substractively to get rid of all the cosmic rays noise. That kind of a scheme is what i was talking about in the first post.


.

 

[cjp]

If you only need to remove the noise by averaging, you can layer the images in photoshop with transparency (Bottom layer 100%, next layer 50%, 3rd layer 25%, 4th layer 12%, and so on).

Are you sure it shouldn't be 100%, 50%, 33%, 25%, 20% and so on?
Suppose you have 4 images: A,B,C,D
Your numbers give:
layer1: 100% A
Adding layer2: 50% layer2 + 50% lower layers = 50% B + 50% A
Adding layer3: 25% layer3 + 75% lower layers = 25% C + 75% * (50% B + 50% A)
= 25% C + 37.5% B + 37.5% A
Adding layer4: 12% layer4 + 88% lower layers = 12% D + 88% (25% C + 37.5% B + 37.5% A)
= 12% D + 22% C + 33% B + 33% A

My numbers give:
layer1: 100% A
Adding layer2: 50% layer2 + 50% lower layers = 50% B + 50% A
Adding layer3: 33% layer3 + 67% lower layers = 33% C + 67% * (50% B + 50% A)
= 33% C + 33.5% B + 33.5% A
Adding layer4: 25% layer4 + 75% lower layers = 25% D + 75% (33% C + 33.5% B + 33.5% A)
= 25% D + 24.75% C + 25.125% B + 25.125% A

Anyway, I once tried the same with some extremely zoomed-in pictures of Jupiter to see if I can find the Jovian moons. I failed, and I figured out later why: to increase my SNR sufficiently, I'd need hundreds of images.

The problem is that the noise only goes down with the square root of the number of images. So, to give you an idea:

# images   Noise (percentage of noise in single image)
1          100%
2          71%
3          58%
4          50%
10         32%
100        10%

You'll not be able to get rid of light pollution by these methods, as the light pollution is an inherent part of the scene that you took the photos of. There are ways to get rid of light pollution. It basically requires taking a copy of the image as a new layer and applying a large gaussian blur to it (that essentially blurs out all stars. Since the light polliution is diffuse anyway, the blur doesn't really affect it so you end up with an image that is pretty much just the light pollution). You can then subtract this blurred image from the original which essentially gets rid of the light pollution. Google some words for better information and, I'm sure, lots of blogs on the subject.

That sounds like a nice method. Basically, it leaves you with all sharp objects, so it's a kind of sharpening filter (the imaging equivalent of a high-pass filter). Unfortunately, noise is also quite sharp, so it gets rid of the overall background light, but you don't get rid of the 'grass'.

 

[Redburne]

Are you sure it shouldn't be 100%, 50%, 33%, 25%, 20% and so on?

I think you're right.