|
|
 |
 |
 |
 |
 |
Everything about image noise and visible sharpness
"Image noise" is the digital equivalent of film grain for analogue cameras. Alternatively, one can think of it as analogous to the subtle background hiss you may hear from your audio system at full volume. For digital images, this noise appears as random speckles on an otherwise smooth surface and can significantly degrade image quality. Although noise often detracts from an image, it is sometimes desirable since it can add an old-fashioned, grainy look to an image which is reminiscent of of early film. Some noise can also increase the apparent sharpness of an image. Noise increases with the sensitivity setting in the camera, length of the exposure, temperature, and even varies amongst different camera models.
Image noise
Some degree of noise is always present in any electronic device that transmits or receives a "signal." For televisions this signal is the broadcast data transmitted over cable or received at the antenna; for digital cameras, the signal is the light which hits the camera sensor. Even though noise is unavoidable, it can become so small relative to the signal that it appears to be nonexistent. The signal to noise ratio (SNR) is a useful and universal way of comparing the relative amounts of signal and noise for any electronic system; high ratios will have very little visible noise whereas the opposite is true for low ratios.
About the digital cameras ISO values
A camera's "ISO setting" or "ISO speed" is a standard which describes its absolute sensitivity to light. ISO settings are usually listed as factors of 2, such as ISO 50, ISO 100 and ISO 200 and can have a wide range of values. Higher numbers represent greater sensitivity and the ratio of two ISO numbers represents their relative sensitivity, meaning a photo at ISO 200 will take half as long to reach the same level of exposure as one taken at ISO 100 (all other settings being equal). ISO speed is analogous to ASA speed for different films, however a single digital camera can capture images at several different ISO speeds. This is accomplished by amplifying the image signal in the camera, however this also amplifies noise and so higher ISO speeds will produce progressively more noise.
Types of Noise
Digital cameras produce three common types of noise: random noise, "fixed pattern" noise, and banding noise. The three qualitative examples below show pronounced and isolating cases for each type of noise against an ordinarily smooth grey background.
 IMAGE of Three noise Types
- Random noise is characterized by intensity and color fluctuations above and below the actual image intensity. There will always be some random noise at any exposure length and it is most influenced by ISO speed. The pattern of random noise changes even if the exposure settings are identical.
- Fixed pattern noise includes what are called "hot pixels," which are defined as such when a pixel's intensity far surpasses that of the ambient random noise fluctuations. Fixed pattern noise generally appears in very long exposures and is exacerbated by higher temperatures. Fixed pattern noise is unique in that it will show almost the same distribution of hot pixels if taken under the same conditions (temperature, length of exposure, ISO speed).
- Banding noise is highly camera-dependent, and is noise which is introduced by the camera when it reads data from the digital sensor. Banding noise is most visible at high ISO speeds and in the shadows, or when an image has been excessively brightened. Banding noise can also increase for certain white balances, depending on camera model.
Although fixed pattern noise appears more objectionable, it is usually easier to remove since it is repeatable. A camera's internal electronics just has to know the pattern and it can subtract this noise away to reveal the true image. Fixed pattern noise is much less of a problem than random noise in the latest generation of digital cameras, however even the slightest amount can be more distracting than random noise.
The less objectionable random noise is usually much more difficult to remove without degrading the image. Computers have a difficult time discerning random noise from fine texture patterns such as those occurring in dirt or foliage, so if you remove the random noise you often end up removing these textures as well. Programs such as Neat Image, Helicon and Noise Ninja can be remarkably good at reducing noise while still retaining actual image information. Please also see my section on image averaging for another technique to reduce noise.
Understanding the noise characteristics of a digital camera will help you know how noise will influence your photographs. The following sections discuss the tonal variation of noise, "chroma" and luminance noise, and the frequency and magnitude of image noise. Examples of noise variation based on ISO and color channel are also shown for three different digital cameras.
Characteristics of Noise in digital sensors
Noise not only changes depending on exposure setting and camera model, but it can also vary within an individual image. For digital cameras, darker regions will contain more noise than the brighter regions; with film the inverse is true.
IMAGE 4 Greytones in RL with corresponding fragments in 100 percent scaled.
Note how noise becomes less pronounced as the tones become brighter. Brighter regions have a stronger signal due to more light, resulting in a higher overall SNR. This means that images which are underexposed will have more visible noise-- even if you brighten them up to a more natural level afterwards. On the other hand, overexposed images will have less noise and can actually be advantageous, assuming that you can darken them later and that no region has become solid white where there should be texture (see "Understanding Histograms, Part 1").
Noise is also composed of two elements: fluctuations in color and luminance. Color or "chroma" noise is usually more unnatural in appearance and can render images unusable if not kept under control. The example below shows noise on what was originally a neutral grey patch, along with the separate effects of chroma and luminance noise.
IMAGE of Noise split into Luma and Chroma noise
The relative amount of chroma and luminance noise can vary significantly from one camera model to another. Noise reduction software can be used to selectively reduce both chroma and luminance noise, however complete elimination of luminance noise can result in unnatural or "plasticy" looking images.
Noise fluctuations can also vary in both their magnitude and spatial frequency, although spatial frequency is often a neglected characteristic. The term "fine-grained" was used frequently with film to describe noise whose fluctuations occur over short distances, which is the same as having a high spatial frequency. The example below shows how the spatial frequency can change the appearance of noise.
 | |  | | |
|
|
Low Frequency Noise
(Coarser Texture)
Standard Deviation of 11.7 |
High Frequency Noise
(Finer Texture)
Standard Deviation of 12.5 |
| |  | |  |
If the two patches above were compared based solely on the magnitude of their fluctuations (as is done in most camera reviews), then the patch on the right would seem to have higher noise. Upon visual inspection, the patch on the right actually appears to be much less noisy than the patch on the left. This is due entirely to the spatial frequency of noise in each patch.
Even though noise's spatial frequency is under emphasized, its magnitude still has a very prominent effect. The next example shows two patches which have different standard deviations, but the same spatial frequency.
| | | | |
|
|
Low Magnitude Noise
(Smoother Texture)
Standard Deviation of 11.7 |
High Magnitude Noise
(Rougher Texture)
Standard Deviation of 20.8 |
| | | | |
Note how the patch on the left appears much smoother than the patch on the right. High magnitude noise can overpower fine textures such as fabric or foliage, and can be more difficult to remove without over softening the image. The magnitude of noise is usually described based on a statistical measure called the "standard deviation," which quantifies the typical variation a pixel will have from its "true" value. This concept can also be understood by looking at the histogram for each patch:
IMAGE
If each of the above patches had zero noise, all pixels would be in a single line located at the mean. As noise levels increase, so does the width of this histogram. We present this for the RGB histogram, although the same comparison can also be made for the luminosity and individual color histograms. For more information on types of histograms, please see:
|
|
 |
|
 |
|
|
|
|
