Noise reduction is the process of reducing the random grain and color speckle that degrade an image, most visibly in shadows and at high ISO. The underlying noise comes from the physics of capturing light, and while it can never be eliminated entirely, modern tools suppress it well enough that very high sensitivities remain usable.
Noise reduction is split into two jobs that behave differently. Luminance noise reduction smooths the brightness grain that looks like film texture, and pushing it too far costs fine detail and gives a soft, waxy look. Color noise reduction removes the red, green, and blue blotches that appear in dark areas, and because the eye is far less sensitive to color speckle than to detail, it can usually be applied generously with little visible loss.
Good software pairs these with a detail or contrast slider that tries to recover edge crispness lost to smoothing, so the practical workflow is to apply just enough luminance reduction to calm the grain, push color reduction as needed, then add back a little detail. Reduction can happen in several places. Cameras apply it automatically to JPEGs, and offer dedicated modes such as long exposure noise reduction, or LENR, which subtracts a matching dark frame to remove hot pixels from multi-minute exposures.
Working from a RAW file in post gives the most control, since the full sensor data is intact and you decide exactly how much smoothing to trade for detail. A second technique that avoids the trade-off entirely is averaging, stacking several identical frames so the random noise cancels while the real detail reinforces, which astrophotographers rely on to produce clean images.
The biggest recent change is machine-learning denoising. Tools that learn the difference between noise and real detail, including the option covered in our guide to AI noise reduction, clean files far more cleanly than traditional sliders and can rescue images shot several stops underexposed. They have effectively raised the usable ISO ceiling of every camera by a stop or two.
Good practice treats noise reduction as a balance against sharpening, since the two pull in opposite directions, and applies it selectively. Masking the effect onto smooth skies and shadows while protecting detailed areas like eyes and foliage keeps an image clean without making it look artificial. Exposing well in the first place, using techniques such as exposing to the right, remains the best noise reduction of all.
A sensible default workflow is to leave in-camera noise reduction modest if you shoot raw, since it does not affect the raw data anyway, then judge each image on its merits in post. Apply only as much smoothing as the final size demands, because a web image and a large print tolerate very different amounts before the loss of detail shows.