Read noise is the random electronic noise that a camera sensor and its readout circuitry add to every pixel value when the collected charge is converted, amplified and digitized. In photography it is a central factor of image quality, because it sets the noise floor that determines how much detail survives in the darkest parts of an image.
Read noise is usually expressed in electrons (e-) RMS, referred to the input of the sensor. Modern sensors reach values of roughly 1 to 5 electrons depending on design and gain setting, which is small compared with bright signals but decisive in deep shadows and very low light.
Where read noise comes from
Read noise arises in the signal chain after light has been collected. Each pixel stores photoelectrons, which a readout transistor converts into a voltage. That voltage is amplified, possibly passed through further gain stages, and finally converted to a number by an analog-to-digital converter (ADC). Every stage adds a small random fluctuation.
Unlike shot noise, which comes from the random arrival of photons and grows with the signal, read noise is essentially independent of the amount of light. It is present even in a frame taken with the lens cap on. The table lists its main components.
| Component | Origin | Typical behavior |
|---|---|---|
| Reset or kTC noise | Resetting the pixel capacitor | Largely removed by correlated double sampling |
| Source follower noise | Readout transistor in each pixel | Often the dominant term in modern pixels |
| Amplifier noise | Column and programmable gain amplifiers | Reduced by higher gain early in the chain |
| Quantization noise | Rounding to a finite number of ADC levels | Small with 12-bit or 14-bit converters |
| Pattern noise | Fixed offsets between rows or columns | Visible as banding in lifted shadows |
How read noise shapes dynamic range and signal-to-noise ratio
Read noise defines the lower end of dynamic range. Engineering dynamic range is commonly calculated as the full well capacity divided by the read noise, expressed in stops. A pixel that holds 50,000 electrons and has 3 electrons of read noise offers about 14 stops, while the same pixel with 12 electrons offers only about 12 stops.
The total noise in a pixel combines read noise and shot noise as the square root of the sum of their squares. In bright areas shot noise dominates, so read noise hardly matters. In dark areas the signal is small, read noise becomes the dominant noise source, and the signal-to-noise ratio drops sharply. The table illustrates this relationship.
| Signal in electrons | Shot noise | Total noise | Dominant source |
|---|---|---|---|
| 10,000 | 100 e- | About 100 e- | Shot noise |
| 1,000 | About 32 e- | About 32 e- | Shot noise |
| 100 | 10 e- | About 10.4 e- | Shot noise, read noise noticeable |
| 10 | About 3.2 e- | About 4.4 e- | Both comparable |
| 2 | About 1.4 e- | About 3.3 e- | Read noise |
Read noise, ISO and ISO invariance
Read noise referred to the input usually falls as ISO rises, because analog gain amplifies the signal before the noisier later stages of the chain. Many sensors also use dual conversion gain, switching to a high-gain pixel mode above a certain ISO, which produces a distinct drop in read noise at that point.
When read noise is already very low at base ISO, raising ISO in the camera brings little benefit over brightening the image later. Such sensors are called ISO-invariant across that range. The table compares typical behaviors.
| Sensor behavior | Read noise at base ISO | Read noise at high ISO | Practical consequence |
|---|---|---|---|
| ISO-invariant | Low, about 2 to 4 e- | Similar or slightly lower | Underexposing to protect highlights costs little |
| Dual gain | Moderate | Drops sharply above the switch point | Use the higher gain step in low light |
| ISO-variant (older designs) | High, often 10 e- or more | Much lower | Raise ISO in camera rather than brighten later |
Pixel size also plays a role. Smaller pixels often have lower read noise per pixel, but when images are compared at the same output size, the results depend on the full sensor design, so pixel pitch alone does not predict the outcome.
Practical choices to limit read noise
The most effective way to reduce the influence of read noise is to collect more light, so that shot noise rather than read noise dominates the shadows. Camera settings then help keep the readout as clean as possible.
- Expose as generously as highlights allow, since more signal pushes shadows above the read noise floor.
- Use the higher gain step on dual-gain sensors in low light instead of brightening a base-ISO file.
- Prefer 14-bit raw where available, because some cameras use quieter readout modes at higher bit depth.
- Stack multiple frames in astrophotography; averaging reduces random noise by the square root of the number of frames.
- Use dark frames or bias frames to remove fixed offsets, although random read noise cannot be subtracted.
Noise reduction in processing can soften visible read noise but also smooths fine detail, so it is a remedy rather than a substitute for adequate exposure.
Misconceptions about read noise
Read noise is often confused with other forms of noise because they appear together in the same image. Separating them clarifies which remedy applies.
| Term | What it describes | Relation to read noise |
|---|---|---|
| Shot noise | Randomness of photon arrival | Independent source, dominates in bright areas |
| Luminance noise | Brightness grain in the image | Visible result that includes read noise |
| Chromatic noise | Color speckles | Often strongest where read noise dominates |
| Thermal or dark current noise | Charge generated by heat | Grows with exposure time, separate from readout |
| Fixed pattern noise | Repeating offsets | Systematic part of readout, partly correctable |
Another misconception holds that higher ISO creates noise. In reality, high ISO is used because there is little light; the noise comes mainly from the weak signal itself, while gain often lowers the relative read noise.
Stripped to the essentials
Every time a camera reads the light values from its sensor, the electronics add a tiny amount of random hiss, much like a quiet background hum in a speaker. That hiss is read noise.
In bright parts of a photo it is drowned out. In very dark parts, where there is little light to record, it becomes visible as grain and blotches. More light on the sensor is the best way to keep it hidden.
Where it is applied
An astrophotographer captures the Milky Way from a dark site with a full frame camera, a 20 mm lens at f/1.8 and a 15-second exposure. The camera has a dual-gain sensor whose read noise drops markedly above ISO 640.
Test frames at ISO 400 brightened later show blotchy, banded shadows, while frames at ISO 1600 look much cleaner at the same exposure time. The photographer then takes 20 frames and stacks them, which reduces the random noise by a factor of about 4.5. The final image shows faint dust lanes that were lost in the read noise of a single base-ISO frame.