Home > Digital Photography Tutorials > Noise Averaging

Noise Averaging

February 1st, 2008

There are several very effective methods for reducing and even eliminating digital sensor noise. I’ve been testing and utilizing these methods with some astonishing results. The image quality I’m producing at 3200 ISO is equivalent to, if not better than 100 ISO.

The techniques I’m using do require multiple images that align precisely. This usually means a tripod and static subject are required; however I do sometimes shoot hand-held “bursts” of images. Photoshop CS3 has excellent aligning technology that comes into play here, as these images must be aligned precisely. When shooting aligned image sets; use manual focus and a fixed aperture. Even slight changes degraded image alignment.

Most sensor noise is random which means it’s in a different place on each exposure. By layering or merging images on top of each other this noise can be averaged out. The more images, the better the average and the better the file. Each individual image may be extremely noisy, but the noise is never in the same place. A noisy (bad) pixel can be negated by stacking good pixels on top of it and averaging them out. 

When is enough, enough? The more images you use, the better the results as long as they line up. How many images you use is really just a work-flow issue. The math for calculating the reduction in noise is not simple. It’s not linear and it’s different for every pixel value, so it gets complicated fast. I’ll leave that to the mathematicians, physicist, and astronomers.

Adobe Photoshop CS3 Noise Averaging Methods:

Click the image for a better look.

Method#1- Layering with Opacity: This method is a nice place to start, though not as powerful as the other methods, there is merit in its simplicity and there are a multitude of options with layers
1) Shoot multiple images that will align, preferably RAW format.
2) Stack the images on layers in a single file. This can be accomplished by dragging the background from one to the other while holding down the shift key. Better yet, is Dr. Browns Place-A-Matic Adobe Bridge Script which will layer multiple files from Adobe Bridge.
3) Highlight all the layers by shift clicking on them; then select Edit/Auto Align Layers.
4) Adjust the opacity of each layer to be the opacity to be the reciprocal of the layer number, i.e. the background layer at 100%, layer 2 at 50%, layer 3 at 33%, etc. If you have 10 images combined each will contribute 10% by using this method.
5) When your ready, flatten the image.

Method #2- 32 Bit Merging : This method is essentially the same as Merging HDR images. The difference is that the merge isn’t necessarily to increase dynamic range. Merging exposure bracketed images will improve both noise and dynamic range, a great one two punch.
1) Shoot multiple images that will align, preferably RAW format.
2) Select them in Bridge and edit them in Adobe Camera RAW (see ACR HDR Settings).
3) From Bridge first highlight the images to be merged; then select Tools/Photoshop/Merge to HDR.
(Note: The Automatically Align Source Images feature is turned on in Photoshop under File/Automate/Merge to HDR)
4) Say OK to the Merge to HDR preview dialog, nothing to do here, it’s a useless stop.
5) Convert the 32bit image to 16 or 8 Bit using Exposure Gamma or Local Adaptation. This is the tough part, mapping the 32 image down to a lower bit depth is tricky. I recommend Photomatix Pro as an alternative method of both merging and tonal mapping 32 bit images.

Method#3- Apply Image : This method is hard core and tedious, but it’s the most powerful. It allows for not only the reduction of noise, but a buildup of signal. By applying an image to another image the signal is multiplied.
Start To Finish
1) Open all images in 16Bit, then convert each to 32bit.
2) Choose an image that others will be applied to, then Select Image/ Apply Image:
>Source will be the image you are applying from. This will be changed each time.
>Layer is what you’re applying to, probably the Background.
>Channel is RGB
>Blending is Linear Dodge (ADD)
>Opacity is 100%
3) Repeat Apply Image selecting a different Source file for each image.
4) Convert the 32bit image to 16 or 8 Bit using Exposure Gamma or Local Adaptation.

These methods obviously don’t work for all situations; the camera and subject need to be stationary, and that’s only possible with a small percentage of photography. When you can shoot on a tripod and your subject isn’t moving, incredibly clean files are possible at any ISO. Noise averaging really does open up a world of photographic possibilities.

Categories: Digital Photography Tutorials Tags:
  1. ron
    February 1st, 2008 at 17:16 | #1

    For the opacity technique, if you want each layer to contribute equally, I believe you’d want the opacity to be the reciprocal of the layer number, i.e. the background layer at 100%, layer 2 at 50%, layer 3 at 33%, etc.

    Or am I missing something?

  2. Michael Stewart
    February 1st, 2008 at 19:24 | #2

    It is I who was missing something, thanks for the correction. I really appreciate the comment and I have corrected the post.

  3. Michael Voigt
    February 7th, 2008 at 21:35 | #3

    This is an amazing technique… I cannot wait to try it out. Will it help with low light situations because you can bump your ISO up?

  4. Michael Stewart
    February 7th, 2008 at 22:42 | #4

    I’m getting amazing files from 3200 ISO. I have some other techniques brewing with Photoshop CS3 Extended.

  5. Ben
    February 8th, 2008 at 17:29 | #5

    Is there any reason you did not try the Median Layer Stack mode in Photoshop CS3 Extended?

    Steps would be to use Dr. Brown’s Stack-a-matic and set everything up in there to keep it simple and fast. You can download that at russellbrown.com


  6. Michael Stewart
    February 9th, 2008 at 21:20 | #6

    Stacks in Photoshop Extended is defiantly the way to go. I intend to update this post with more the info from my Extended research, but not everyone is using extended. The HDR method is easy.

Comments are closed.