Excellent article on reducing high-ISO noise

[Scottes]

http://www.photo.net/learn/dark_noise/

[Radtech1]

Boy, is my face ever red! :oops:

I cannot believe that I did not think this on my own. Not that should be intuitively obvious to just anybody however I use this technique every single day at work.

I work in the field of Nuclear Medicine, which is kind of like x-rays from the inside out. The patient will be the source of the radiation, and the camera is nothing but a huge receptor. The photomultiplier tubes in the camera fluctuate in their response over time. About once a month we just open a radioactive source and leave the room for a couple of hours. This allows every single pixel to get completely saturated. Any variation in response will be seen. On every scan we do, we invert that saturation image, and apply it to the images of the patients. The term that we use to refer do it is a "correction flood field".

But I do have one question, the author specifically indicates shooting in raw format. I wonder if there is a workaround for those of us to shoot JPEG?

Rad

[PacAce]

Boy, is my face ever red! :oops:

I cannot believe that I did not think this on my own. Not that should be intuitively obvious to just anybody however I use this technique every single day at work.

I work in the field of Nuclear Medicine, which is kind of like x-rays from the inside out. The patient will be the source of the radiation, and the camera is nothing but a huge receptor. The photomultiplier tubes in the camera fluctuate in their response over time. About once a month we just open a radioactive source and leave the room for a couple of hours. This allows every single pixel to get completely saturated. Any variation in response will be seen. On every scan we do, we invert that saturation image, and apply it to the images of the patients. The term that we use to refer do it is a "correction flood field".

But I do have one question, the author specifically indicates shooting in raw format. I wonder if there is a workaround for those of us to shoot JPEG?

Rad

Yes, there is. Shoot RAW! :mrgreen:

Scott, interesting article. I've read bits and piece about this "calibration" process but this one really covered the subject matter a little more in detail. Thanks.

[Chazs]

This sounded promising, but after four different attempts I just couldn't get it to work. Each time the process increased the "noise" tremendously, especially in the blues. I also tested 4 different dark frames (at the same exposure) and came to the conclusion that the noise I get at 3200ISO is truely random and not predictable. These were all done at about 1/4 second at f16. Maybe it works better at longer exposures. I'll keep trying.

[F1_Fan]

Dark frame works better for hot pixels than it does for noise. Noise is random so DFS will work better sometimes and worse others. I've used DFS for years (I learned digital imaging by using astronomical CCDs in the early 90's where this technique was essential).

Here's my example done in about 1997 with a Kodak DC260... a horrible little camera when it came to long exposures.

http://members.shaw.ca/digitallattice/darkframe.html

[Scottes]

This sounded promising, but after four different attempts I just couldn't get it to work.

If you produce the two "filter" frames and then subtract one from the other you should get an almost black image. Are you using RAW?

Note that I have not tried this yet....

[maderito]

This sounded promising, but after four different attempts I just couldn't get it to work. Each time the process increased the "noise" tremendously, especially in the blues. I also tested 4 different dark frames (at the same exposure) and came to the conclusion that the noise I get at 3200ISO is truely random and not predictable. These were all done at about 1/4 second at f16. Maybe it works better at longer exposures. I'll keep trying. Ditto - except I stopped after 2 attempts with ISO 3200, larger aperture and faster shutter speed. Noise reduction with Neat Image and Noise Ninja gave good results (virtually identical). These programs work best on random noise.

[Scottes]

After the comments by Chazs, Maderito and F1 I have to say that I'm no longer inclined to try this. Thanks for saving me some time guys!

:( It certainly sounded pretty cool though.

[Jesper]

This is dark frame substraction, exactly what the 20D (and other cameras that have that feature) does when you turn on the noise reduction feature.

Interesting to know how you can do it by hand, because the 10D doesn't have the 20D's noise reduction.

[PacAce]

This is dark frame substraction, exactly what the 20D (and other cameras that have that feature) does when you turn on the noise reduction feature.

Interesting to know how you can do it by hand, because the 10D doesn't have the 20D's noise reduction.

Yup, that's what I thought, too, but for the life of me, I can't get a good noise-free image out of the 10D using this technique. And I've tried many times. Actually, the images came out a little worse after this process was applied. I'm guessing that the 10D at 1600 is so good compared to other camera that noise (the kind this process is supposed to eliminate) is not an issue with the 10D. Maybe at longer exposures, this technique may work but I haven't tried any long exposures. I've only gone as high as 30 seconds and still couldn't get the process to work.

BTW, an interesting observaton I've made while trying out this process. If you use "Exclusion" instead of "Difference" and watch the histogram, you'll see something interesting. It seems to remove some noise but it leaves other noise data along so that you end up with spikes on the histogram curve from the noise that wasn't touched. I haven't had time to research what the "Exclusion" blending option does so if anybody knows, just chime in here. :)

[CyberDyneSystems]

Awesome article Scottes!

Rad,.. lol,. you should be redfaced! :)

A buddy of mine who knows nothing of any of this.. but is a clever CAD professional got his first digital camera a few years back (an olympus P&S) and within about two weeks had made a dark frame to fix hot pixels!

It is a great article.. and I never would have thought that it would work for ISO noise as well.. as I never would have thought that ISO noise is NOT random.

So.. is he wrong in saying it's not random?
What on earth was the author doing that made it work for him then?

Very interesting indeed :)

If this did work... It seems that one could actually make a series of "dark frames" at various ISO settings and exposure times as well.. and keep a library. It would be much more hit and miss and not as accurate.. but for those times when you did not plan ahead ;)

[PacAce]

Awesome article Scottes!

Rad,.. lol,. you should be redfaced! :)

A buddy of mine who knows nothing of any of this.. but is a clever CAD professional got his first digital camera a few years back (an olympus P&S) and within about two weeks had made a dark frame to fix hot pixels!

It is a great article.. and I never would have thought that it would work for ISO noise as well.. as I never would have thought that ISO noise is NOT random.

So.. is he wrong in saying it's not random?
What on earth was the author doing that made it work for him then?

Very interesting indeed :)

If this did work... It seems that one could actually make a series of "dark frames" at various ISO settings and exposure times as well.. and keep a library. It would be much more hit and miss and not as accurate.. but for those times when you did not plan ahead ;)

Don't forget, dark noise is also dependent on temperature so you'd have to make temperature another criteria for indexing the dark noise library. :)

[Scottes]

Don't forget, dark noise is also dependent on temperature so you'd have to make temperature another criteria for indexing the dark noise library.
So let's see... temperature, exposure time, aperture... Seems like this would only be wrothwhile *sometimes* since it's a bit of work.

If it worked, that is...

[CyberDyneSystems]

I forgot temperature.. looks like it's getting to be a rather huge library !