What is noise?

Noise is the digital equivalent of film grain. It can even look like grain, though more often it looks more like ugly speckles or color artifacts. It results from a variety of sources, including sampling errors in pixels, temperature-induced "dark current" in sensor elements, and signal amplification circuits.

Just as high-speed film yields more grain than low-speed film, digital photos taken at high ISO settings show much more noise than photos taken at low ISO settings. Nearly all compact digital cameras show obvious noise at ISO 400 or above. Even top-of-the-line digital SLRs are susceptible to noise, particularly at high ISO settings.

Film scanners are also well known for introducing noise into digitized images, especially in dark areas of slides and in the blue channel.

Noise is an inherent property of digital imaging sensors. The laws of physics make it impossible to completely eliminate noise, and they force a tradeoff between noise levels and other properties like sensor size or sensitivity. Photons, for instance, arrive at random intervals, so the simple task of counting them during an exposure-- which is the basic function of a pixel in a sensor -- is subject to sampling error. When the exposure is shortened or the pixel size is reduced, there are fewer photons to "average out" the sampling error, so the noise increases relative to the signal.

The small sensors in compact digital cameras are more prone to noise than the large sensors used for digital SLRs. Compact digicams often have as many pixels as their DSLR brethren, but those pixels are packed into one quarter the space -- or even less. So, for any given exposure, many fewer photons reach each pixel in the smaller sensor than in the larger one, and this leads to correspondingly higher noise. So, the noise in a compact camera at ISO 200 might be the same as the noise in a DSLR at ISO 800. By the same reasoning, an 8-megapixel camera might have much higher noise levels than a 4-megapixel camera if both have the same sensor size.

The problem with noise

Many common photography situations (for instance, fast-action sports, indoor, and low-light outdoor photography) can require high ISO settings to avoid motion blur or handshake. Without the ability to control noise, the photographer is faced with a choice between two bad alternatives: Use a low ISO and get a blurry photo, or use a high ISO and get a noisy image.

In addition, photographers who make large prints often notice noise in smooth areas even for images taken at low ISO settings. While this isn't a problem for someone who only makes 4"x6" prints, it is an issue for the professional who must frequently create poster-size enlargements from today's partial-frame DSLRs.

In both situations, noise removal is desirable to increase the visual quality of the image. Unfortunately, digital camera noise is very difficult to remove using conventional image editing software:

Camera noise is spread across the frequency spectrum. It includes "fine-grained" components as well as "coarse" components.


Noise varies with color and brightness, and it is different for every camera and scanner. For instance, blue-channel noise is often higher than in other channels, and shadow noise is usually higher than in bright areas.

Most commercially available noise removal tools fail along one or both of these dimensions. Typically, they are based on ad-hoc methods like adaptive median filtering, thresholding, or photo-editor macros, so they are inherently restricted to a limited frequency range, and they generally assume that noise is uniform throughout the image, or they rely on a limited set of parameters for each image. So, they tend to work well on certain images that "fit" their methods well. However, they are not robust, and they tend to yield poor results when presented with a variety of images.