Signal-to-noise ratio (SNR) is the ratio between the useful image signal and the random noise that accompanies it, and it is the main measure of how clean a digital photograph looks. In image quality for photography, SNR is evaluated per pixel or per area of the image, and a higher ratio means smoother tones, finer detail and less visible grain.
SNR is usually expressed as a plain ratio, in decibels (dB), or in stops (powers of two). It depends above all on how much light reaches the sensor, which makes it the link between exposure decisions, sensor size and the noise seen in the final image.
How signal and noise arise in a sensor
The signal is the number of photoelectrons each pixel collects during exposure, which is proportional to the light that falls on it. The noise is the random variation around that value from frame to frame or pixel to pixel. Several independent noise sources add together in quadrature, as the square root of the sum of their squares.
| Noise source | Origin | When it dominates |
|---|---|---|
| Shot noise | Random arrival of photons | Midtones and highlights |
| Read noise | Readout electronics and amplifiers | Deep shadows and very low light |
| Dark current noise | Thermally generated electrons | Long exposures and hot sensors |
| Pattern noise | Fixed pixel and column differences | Heavily lifted shadows |
| Quantization noise | Rounding to digital values | Rarely visible with 12 to 14 bit data |
Shot noise follows Poisson statistics, so its magnitude equals the square root of the signal. This leads to a fundamental rule: in the shot-noise-limited regime, SNR equals the square root of the number of collected photons. Collecting four times more light doubles SNR and improves it by one stop.
Measuring and expressing SNR
SNR can be stated in several equivalent ways. Engineers often use decibels, where 20 times the base-10 logarithm of the ratio is used for amplitude signals, while photographers more often think in stops, where each stop doubles the ratio.
| Ratio | Decibels | Stops | Typical visual impression |
|---|---|---|---|
| 1:1 | 0 dB | 0 | Signal lost in noise |
| 4:1 | About 12 dB | 2 | Very noisy, barely usable |
| 10:1 | 20 dB | About 3.3 | Usable shadows, visible grain |
| 40:1 | About 32 dB | About 5.3 | Clean to most viewers |
| 100:1 | 40 dB | About 6.6 | Very clean midtones |
Standards such as ISO 15739 define procedures for measuring noise and SNR in digital still cameras. Comparisons between cameras are only meaningful when they use the same output size, because downsampling an image averages pixels and raises per-pixel SNR.
What determines SNR in practice
The most important factor is total light captured, set by the aperture, the shutter speed and the scene brightness, and by the sensor area that collects it. ISO itself does not add light; it amplifies the signal and the noise already present, although higher analog gain can slightly reduce the relative contribution of read noise.
| Factor | Effect on SNR | Reason |
|---|---|---|
| Wider aperture | Higher | More photons per unit area |
| Longer shutter time | Higher, until motion blur or heat | More photons collected |
| Larger sensor at equal output size | Higher | More total light for the same framing |
| Raising ISO in the same light | Roughly unchanged or lower | No extra light, gain amplifies noise |
| Underexposing and brightening later | Lower | Less light captured in the first place |
| Averaging several frames | Higher by the square root of the frame count | Random noise partially cancels |
Pixel size matters less than often assumed. Smaller pixels have lower per-pixel SNR, but at the same output size and sensor area, the total light is the same, so images from high-resolution and low-resolution sensors of equal size show similar noise when viewed equally large.
SNR compared with related image quality measures
SNR is closely tied to dynamic range, which is the ratio between the brightest recordable signal and the noise floor. Dynamic range is effectively the span of exposure levels over which SNR stays above an acceptable threshold.
- Dynamic range describes the usable tonal span; SNR describes quality at a given tone.
- Luminance noise and chromatic noise are visible forms of low SNR, in brightness and color respectively.
- Sharpness is separate, but aggressive noise reduction to raise apparent SNR can destroy fine detail.
- Read noise sets the SNR floor in shadows, while shot noise sets it in brighter tones.
Improving SNR and avoiding common errors
The most effective way to improve SNR is to capture more light: expose to the right without clipping important highlights, use a wider aperture or longer exposure when the subject allows, and use image stacking for static scenes such as night landscapes.
A common error is to keep ISO low at all costs in dark scenes and then brighten the file heavily. With many sensors this gives worse shadow noise than setting a higher ISO in camera, because the brightened file inherits the full read noise of a low-gain readout.
The simple explanation
Signal-to-noise ratio compares the picture information a camera records with the random speckles that come with it. Imagine listening to a voice on an old radio: the voice is the signal, and the hiss is the noise.
When there is plenty of light, the voice is loud and the hiss is hard to notice. In darkness, the voice is quiet and the hiss takes over. More light almost always means a cleaner photo.
How it is used in practice
An astrophotographer photographs the Milky Way from a dark site with a full frame camera, a 20 mm f/1.8 lens and 15 second exposures at ISO 3200. A single frame shows the galaxy clearly but with heavy grain in the dark sky.
She captures 16 identical frames and averages them in stacking software. Because random noise partly cancels, SNR rises by the square root of 16, a factor of four or two stops. The stacked image shows smooth sky gradients and faint dust lanes that were hidden in noise before.