« Nikon D200 Zone System Dynamic Range | Main | Nikon D200 Black and White Photography »
March 19, 2006
High ISO Noise Reduction by Image Averaging
If you are photographing a stationary subject with a stationary camera you can reduce grain or high ISO noise by averaging together multiple seemingly identical images.
I set up a still life under relatively low light conditions and photographed it eight times in a row at 1600 ISO with a Nikon D200 mounted on a tripod.
Here is a 100% crop of the 1600 ISO image. It's noisier than I'd like it to be.
I have now taken all 8 images and stacked them up in Photoshop CS2 in layers 0 through 7. Let's see what happens when we average them together.
First I set the opacity of layers 7, 5, 3, and 1 to 50% Next I merge them down in pairs -- 7 to 6, 5 to 4, 3 to 2, and 1 to 0. I now have 4 layers in a stack labelled 6, 4, 2, and 0
Here's what a pair of layers looks like when averaged together. You can already see a noise reduction.
Let's continue, repeating the process.
I'm now setting layers 6 and 2 to 50% and merging down in pairs -- 6 to 4, and 2 to 0. I now have 2 layers in a stack labelled 4 and 0.
Let's repeat the process one final time.
I set layer 4 to 50% and merge it down onto 0.
Look above. Look below. You can really see a difference.
Here's the original noise again -- just for comparison.
You can do grain reduction through image averaging on moving footage as well, just as long as the subject and camera are still. That may not sound useful, but the technique does have its place:
I had occasion to use this technique in the feature film Independence Day. There's a shot of The Empire State Building down at the end of an avenue waiting to explode. The model miniature footage was excessively grainy. I had the compositor replace every frame of the footage with a running average of the previous 20 frames. It made for a great grain removal technique. Once the model exploded, we returned to the grainy element, but it was so mixed with fire and flying debris that the grain wasn't objectionable.
Fun Idea
If you can steady the frames properly, you can use image averaging to get a better look at that flying saucer footage, or you can demosaic the identity- or nudity-obscuring blocky pixel mosaics sometimes placed on running video. The trick is to be able to get the frames to stack up in perfect register, which may not always be possible.
Additional Links
Reindeer graphics Image Averaging
Posted by digital artform at March 19, 2006 07:23 PM
Comments
Couldn't you do this same technique of reducing noise by duplicating one image 8x then using each one as a separate layer?
t
Posted by: troy at March 24, 2006 02:01 PM
I don't think so. The noise has to vary randomly. When you average those random fluctuations you approach the ideal mean from which they deviated.
There are, however other ways of doing the averaging in Photoshop. One popular way involves setting the opacity of a given layer to 1/(N+1) where N = # of layers below the given layer in the stack.
An example of this equation in action:
http://www.cambridgeincolour.com/tutorials/noise-reduction.htm
I enjoyed the photography on your web site, btw.
Posted by: Joseph Francis at March 24, 2006 03:27 PM
T,
That's right, it wouldn't work with a single image because the noise pattern would be identical for every image, so you'be in essence be averaging nothing. The point of taking 8 shots is that the noise is more or less random from shot to shot. Therefore, when you average the noise, the noise cancels out vis a vis the true signal from the underlying image detail.
Posted by: Kenn at April 13, 2006 06:01 PM
Fascinating technique. I was mulling it over, and in the context of still photography, I'm not sure exactly when it would be preferable over a longer shot at lower ISO. Perhaps if you had no tripod and could rattle off a 3-5fps burst, and didn't mind cropping to compensate for panning/rotation adjustment during the aligning process?
Posted by: Kenn at April 13, 2006 06:30 PM
Moreover, I wonder what would happen if you combined a few dozen low ISO photos. Theoretically, you could obtain a perfectly clean image, right?
Posted by: Zach Work at April 20, 2006 10:11 AM
Yes, the more images, the less noise.
Posted by: Joseph Francis at April 20, 2006 02:31 PM