Reducing Digital Camera Noise Using Multiple Exposures
Updated: November 6, 2007
Today’s digital SLRs offer very low noise levels when set to their base ISO (typically ISO 50 or 100). Even so, dark areas tend to show significant noise levels which can impede sharpening and obscure detail. However, multiple exposures can be averaged to nearly eliminate digital noise. This can be done manually, or specialized software such as Photo Acute can be used to automate merging multiple frames for improved sharpness, noise, depth of field, etc. This article explores how well the multiple frame technique works for reducing digital noise.
The well-known digital photography “expose to the right” technique should always be considered (with RAW files) before resorting to a multi-frame approach: “overexpose” the image as much as possible without losing any highlights in the R/G/B channels, then compensate for the overexposure in the RAW-file processing software. (Watch for a future article on this topic at diglloyd.com). The ETTR technique can and should be used in conjunction with the multi-frame approach.
For example, if a +1.5 stop exposure was made in the camera, the RAW software would use -1.5 stop exposure compensation, resulting in the desired +0 “correct exposure. This technique fully exploits all the precision available, pushing very dark pixels that might be represented using only 2 or 4 or 6 bits into 3 or 5 or 7 bits (adding one bit/stop doubles the precision).
While useful as everyday practice, the “expose to the right” technique is of limited benefit and applicability; it is effective only for low contrast images; high-contrast images are already using the full tonal scale and increasing exposure would simply blow out image highlights. Still, I use the ETTR technique for nearly all his tripod-based shots, since there is no real downside other than remembering to compensate during RAW-file conversion.
To obtain high-quality results, care should be taken that the camera remains perfectly fixed in place across all exposures. Vibration or movement of the tripod, ballhead, etc between frames can introduce slight misalignment of images that can blur image detail when they are combined. Even pressing the shutter release can move the camera, so use of a remote release is advised (cabled or wireless, available for most digital SLRs). Make sure all knobs are locked down, and enable the mirror lockup function if your camera offers it. Appropriate technique is no different than that used to obtain the best image quality when a single frame is to be taken (see The Sharpest Image).
I do not recommend multi-frame bursts for making multiple exposures; this will introduce camera motion due to “mirror slap” that can result in multiple pixel jiggles between frames. Take individual frames using mirror lockup, allowing at least 2 seconds between mirror up and the exposure, using a remote release or self-timer.
While specialized software programs might mitigate sloppy technique to some extent, taking the time to do it right is a skill that is invaluable for all types of photography. Bothering with such advanced techniques only makes sense if perfect technical execution is achieved.
The averaging techniques discussed on this page assume identical exposures averaged together. Software-based techniques [1, 2] might involve using multiple exposures at +0EV, +1EV, +2EV, etc. That approach is best done using specialized software to compensate for the exposure difference. It also introduces the risks of non-linear color shifts (which I have observed in Nikon digital SLRs).
It is self-defeating to employ this technique using JPEG files, which introduce a block-pattern noise of their own. Shoot RAW using manual exposure and process all images identically into 16-bit TIF files. Ensure that all frames are taken with the same focus (switch to manual focus after first focusing using your desired method).
Comparisons — caveat
Results on this page from the Nikon D200 and Canon EOS 1D Mark III examples should not be compared directly; lighting, white balance and other adjustments were done differently.
The Nikon D200 offers multiple exposure capability within the camera. With enough exposures, a smooth noise-free image should result. The advantage of this feature is that a single RAW file results, saving space and avoiding the confusion of multiple files. The disadvantage is that technical errors in one frame might degrade the combined image that includes it. An example would be bumping the camera during one of the exposures, or the wind moving tree branches.
The Nikon D200 offers an Auto Gain setting for its multiple exposure feature; it ensures that the final multiple-exposure image matches the brightness to be expected when making a single exposure. That setting was used for this test.
Base exposure was 1.6 seconds at ISO 100. Mirror lockup with an MC-20 remote release was used so as not to disturb the camera position. The image was slightly underexposed, and a “curve” was used to bring up the dark areas for ease of viewing.
The actual-pixels crops shown below do show a notable reduction in speckles (noise) with two frames, and even smoother results with four frames. Ironically, “banding” does not improve and might even be more noticeable; the reduction in random noise reveals abrupt tonal transitions in extremely dark areas.
Click on an image to see it at 200% of actual pixels. The images were saved losslessly as PNG files so there are no artifacts present due to compression.
With the D200, four exposures reduce noise substantially, but six or eight exposures might be beneficial. It also appears that the limitations of 12-bit might be coming into play; tonal transitions appear to be less smooth than with the 14-bit Canon EOS 1D Mark III.
Unlike the Nikon D200, the Canon EOS 1D Mark III does not offer a multiple exposure capability. Instead, the images can be merged using thecontrol on each “layer” in a program like Adobe Photoshop CS3.
Base exposure was 1 second at ISO 100. The 1D Mark III offers an ISO 50 setting which was not used, but might be a better alternative for many subjects, though it tends to alter the tonal curve as compared to ISO 100.
Mirror lockup with a remote release was used so as not to disturb the camera position. The image was slightly underexposed, and a “curve” was used to bring up the dark areas for ease of viewing.
Click on an image to see it at 200% of actual pixels, which will make the improvement much more obvious. The images were saved losslessly as PNG files so there are no artifacts present due to compression.
With the Canon EOS 1D Mark III, two exposures provides noticeable improvement. Four exposures largely eliminates the noise; more than that is likely overkill for most purposes.
The multiple exposure technique is quite effective in reducing noise in dark areas. Even two exposures significantly smooths out random noise, with four exposures showing excellent results. The reader should remember that this technique probably does not make sense at ISO 1600; reduce ISO to ISO 100 or ISO 50 first, then apply the technique if it is warranted for the subject matter.
Is the technique worth the effort? That will depend heavily on the subject matter, the intended processing, the size of enlargement, etc. It seems most applicable to high contrast images, where pulling additional detail out of shadow areas is desirable. Or it might be quite useful when exposures in the 1-30 second range are required, a situation in which built-in noise reduction is not fully effective. It certainly cannot hurt to make more than one exposure, preserving future opportunities, but as an everyday technique it might not be a good return on invested time and effort.
Photographers should expect camera vendors to solve the noise problem in the camera automatically; the techniques discussed on this page are poor substitutes for image processing that can and should be engineered into the camera, and done automatically in a single exposure. There is no technical reason that a camera could not be engineered to take multiple exposures (opening the shutter just once) and combine them together into a high-dynamic-range and near-zero-noise image. Current sensor technology might limit the options, but that’s a short-term problem.