Noise Reduction in Digital Photography

By Steve Anchell Back to

Figure 1. Mel Brown at Christo’s Lounge in Salem. This image was taken towards the end of the set and the noise is substantially worse than at the beginning due to the sensor becoming overheated. Leica M9, 50mm f/2 Summicron lens. ISO 2500, f/4 @ 1/30 second. Figure 1. Mel Brown at Christo’s Lounge in Salem. This image was taken towards the end of the set and the noise is substantially worse than at the beginning due to the sensor becoming overheated. Leica M9, 50mm f/2 Summicron lens. ISO 2500, f/4 @ 1/30 second.

Much has been made of digital noise being similar to grain in film. And like film grain, the photographer needs to decide to use it or lose it, often depending on the image. For example, if the image is a landscape meant for large format reproduction, then noise is a bad thing, with less being better. On the other hand, if the image is a street scene or a music show, then noise can add a gritty, gutsy feeling, one that is lost by eliminating all the noise. Even if you choose to keep the noise it helps to know what it is and what causes it to occur. With a little knowledge you can manipulate noise to your advantage, either accentuating or minimizing its appearance.

It may help to know that noise is not unique to digital photography, but a side effect that occurs with all electronic devices, for example, the background hiss of a radio, or the distorted sound of an over-amplified guitar. In many cases it is accentuated by heat. Heat caused by the rapid movement of electrons across a circuit, and ambient heat from physical surroundings. Both of these have an adverse effect on the quality of digital images and the appearance of noise.

Something else to consider: as the camera sensor ages, noise becomes more pronounced. This means that unlike film cameras, digital cameras should be replaced every few years. A general rule of thumb for high-end cameras is after every 100,000 actuations (shutter releases).

In digital photography although there are several types of noise, three types are the most prevalent. These are color, luminance and hot pixels. Major camera makers address this noise with some degree of success, particularly in high-end cameras.

Figure 2a. Dominoes at Night, Havana. This image was recorded using an ISO 3200 setting on a Leica M9. The exposure was f/5.6 at 1/25 second.

All three are more pronounced when too many pixels are squeezed onto the surface of the camera sensor.

Which brings me to ruminate as to why camera makers are vying to squeeze more pixels on sensors than ever before. There are two reasons, one is because they can and it helps to sell cameras, and the second is because they assume you will be able to minimize noise through the use of software.

They can do the first because photographers have no choice other than to buy their over-pixilated cameras; they can do the second because there is software available to reduce the noise, including a High ISO Noise Reduction (NR) setting in most high-end cameras. The problem with the software solution, including the High ISO NR camera setting, is that eliminating noise almost always results in some degree of image degradation, particularly in the area of sharpness.

Figure 2b. This closeup detail shows a high incident of hot pixels, which appear as white pinholes throughout the image.

As far as the High ISO NR setting, I suggest leaving it turned off. It slows down the write speed of the camera and produces a less sharp image. The NR algorithm it applies can be done with a greater degree of control in Lightroom (LR) or other NR software programs with better results. The following is a brief explanation of the three kinds of noise, which I feel are important photographers to know about.

Luminance noise is an overall effect that is similar to film grain, to which it is often compared. It is evenly distributed across an image and is usually the result of using a high ISO. With most pre-2011 camera models that would be ISO 800 or higher. With newer cameras that would be ISO 1600 or higher.

Color noise appears on the screen or in the print as small red, green, or blue dots and is mostly seen in shadow regions. Long exposures, more than one or two seconds, will often greatly increase the incidence of color noise, as will simple underexposure. Moving the histogram to the right, that is, overexposing, will help to eliminate color noise, though it may result in image blur due to the long exposure time.

Figure 2c. Applying the Lightroom Grain−Medium preset eliminates the hot pixels.

Both luminance and color noise can be controlled in the camera by using wider aperture lenses such as f/1.4, f/1.8, f/2.8, for shorter exposure times, and a lower ISO whenever possible. However, using a large aperture will limit your depth of field, should this be a consideration.

The last of the three, hot pixels, is not as prevalent today as it once was due to software solutions built into newer cameras. Hot pixels appear as white pinholes in an image. There are several causes, but the most common is due to the sensor heating up. One is leaving the camera in the heat, say the front seat of your car where the sun strikes it, or in the trunk. Another is simply photographing too fast, or for sustained periods, which often happens with action photography; electrons passing through the sensor cause it to heat up.

I often photograph musicians, such as drummer Mel Brown for Christo’s Lounge in Salem. I will photograph almost continuously for a full 45 minute set. The first few images will have significantly less noise than those toward the end−in fact, in images made the last twenty minutes the noise almost overpowers the image in areas of fine detail. Figure 1 was taken 23 minutes into a set and shows substantially more noise than those taken earlier.

Figure 3. When using the NR panel in LR, start by eliminating color noise before reducing the overall luminance noise. When the noise is substantially reduced, add sharpening.

As mentioned, there is a simple algorithm used by many camera makers that recognizes hot pixels and replaces them with the most likely color, based on an average of its nearest neighbor pixels. This works very well in most cases, but in extreme cases hot pixels may still appear. A simple fix that works most of the time is to add a light or medium film grain pattern to the image. These are often available as presets in image processing programs such as Adobe Lightroom 4. Figures 2a-2c.

A side benefit to using film grain is that it creates an even pattern across the entire image, helping to improve the appearance of luminance noise, while not eliminating it.

Although eliminating hot pixels is fairly simple, color and luminance noise are problematic. Most, if not all, approaches to noise reduction depend on smoothing the edges of the pixels, resulting in varying degrees of sharpness loss in the image. LR4 has a fairly good noise reduction program, under the Detail panel. Figure 3.

If you use the Noise Reduction program in LR start by reducing the Color Noise first, even though Adobe has placed Luminance on top. Preview the image at 1:1 then slowly move the Color slider to the right until the color pixels disappear, then stop. In order to maintain as much sharpness as possible don’t go any further than you need to. Once the color pixels are gone, they’re gone.

The Color Detail slider controls how the edges will be affected. The default is 50 and I usually don’t like to go much more than that.

Figure 4. Noise correction using DxO Optics Pro 8, including adding micro-contrast to improve the overall result. DxO is a stand-alone program that considers every aspect of image enhancement.

With the color noise minimized move to the Luminance slider to minimize the overall grain effect. Again, slowly move to the right until the luminance noise smoothes out. You may need to leave some luminance noise simply to keep the image from becoming too soft. The Luminance Detail slider works similarly to the Color Detail slider, but you have a little more flexibility to sharpen the image by moving it to the right. Too far though, and the grain effect comes back with a vengeance.

The final slider in the LR Noise Reduction panel is used to increase or decrease the contrast of the image. In a way this is a “touch-up” tool to repair some of the damage done by the other corrections. In a seriously noisy image it can help. But if you go too far to the right the image might start looking “blotchy” in some areas. As with all noise corrections, use judiciously.

Overall the noise reduction program in LR does a pretty good job−better even than the one in Photoshop (PS), though it doesn’t allow local control, as does PS (local control is available in LR through the brush tool, but there is no custom control as there is with the NR panel). But is there a better way to tame noise? Actually, there are several noise reduction programs available. Two good ones are Noiseware Professional and DxO Optics Pro 8.

Figure 5. Noiseware is a plug-in for use inside PS, PS Elements and Apple Aperture. Of the three programs shown it allows the user the most control over noise reduction.

DxO Optics Pro 8 is a stand-alone suite of powerful tools for correcting everything from lens distortion to noise. DxO is exceptional in that the company based their software on acquiring and testing as many camera and lenses as possible and creating profiles based on each individual camera. So when you upload an image for correction the program reads the embedded EXIF metadata file and tailors the corrections to the camera and lens that was used. There is probably no better all-in-one image enhancement or correction program available. Figure 4.

Noiseware is one of three programs developed by Imagenomic. Their other two products are Portraiture: Skin Retouching and Realgrain: Film and Grain. Figure 5.

Noiseware works as a plug-in with PS, PS Elements and Apple Aperture. What makes Noiseware unique is that instead of using the standard method of applying a median filter to the image to subtract noise, Noiseware has created its own proprietary noise-filtering algorithm. This program has the most complete set of controls for reducing noise while maintaining quality and sharpness.

If you are looking for a professional program that contains complete image enhancement tools, even beyond those normally found in image editing programs such as LR, then you should consider DxO Optics Pro 8. If you have PS or Aperture and all you need is noise reduction then Noiseware may be a better choice than purchasing the entire DxO suite.

Resources: Adobe Lightroom 4- adobe.com; DxO Optics Pro 8- dxo.com; Noiseware Professional- imagenomic.com


About the Author

Steve Anchell
00256
Steve Anchell is an internationally published photographer, teacher and writer. His books The Darkroom Cookbook, The Variable Contrast Printing Manual and The Film Developing Cookbook are international photography bestsellers. steveanchell.com