Noise

Noise is the random speckle of brightness and color that makes a photo look grainy, strongest in its dark areas. It matters because it sets how dark a scene you can photograph, how far you can brighten a file, and how large a picture can be shown before it looks rough.

Most people learn that noise comes from high ISO. That is the most common misunderstanding in digital photography. Noise comes mostly from too little light reaching the sensor. A high ISO is usually the sign that the light was short, not the cause of the grain. Once you see noise as a light budget, every way of reducing it, from a wider aperture to a tripod to stacking frames, becomes the same idea: get more light, or average away the randomness.

Noise Is Mostly Missing Light

Light arrives as photons, and photons arrive at random, like raindrops on a pavement. If a pixel should catch 100 photons on average during an exposure, on one frame it may catch 91 and its neighbor 108. That scatter is called shot noise (or photon noise). It belongs to light itself. No sensor, however good, can remove it.

The scatter follows a simple law: it is about the square root of the number of photons caught. A pixel that catches 10,000 photons varies by about 100, a signal to noise ratio (SNR) of 100 to 1. A pixel that catches 100 photons varies by about 10, an SNR of only 10 to 1. The noise grew in absolute terms, but the signal grew much faster. What your eye sees as “noisy” is a low ratio, and a low ratio means few photons.

Two rules fall straight out of the square root:

• Four times the light doubles the SNR. Two stops more light, from any source, makes the noise look half as strong.
• One stop more light improves the SNR by about 1.4 times (the square root of 2). One stop is a visible step; two stops is a big one.

The electronics add a second, smaller kind: read noise, a fixed few photons’ worth of error added each time a pixel is read out, whatever the light. In a well lit area it is swamped by shot noise. It only matters in the deepest shadows, as the table below shows.

Want to see it happen? In the camera simulator, raise the ISO on the Night street scene, press Magnify, then switch the camera to a phone sensor.

Why Noise Lives in the Shadows

Every frame holds many light budgets at once: the bright sky caught plenty of photons, the shaded doorway caught very few. So noise is never even across a photo. Here is one pixel’s SNR at different tones in the same exposure. The numbers are illustrative but realistic: a pixel that holds 40,000 photons before it clips, and read noise of 3 photons. The SNR counts both kinds of noise.

Read the table from top to bottom: each two stops darker, the SNR halves, exactly the square root rule. Only in the bottom two rows does read noise make a visible dent (with shot noise alone those would be about 6.3 and 3.1 to 1). This is why the bottom of the dynamic range is set by noise: the darkest “usable” tone is simply the point where the noise becomes as loud as the detail.

So on a digital sensor, the way to keep noise down is to give the shadows enough light. That is the whole logic behind exposing to the right: give the sensor as much light as the highlights allow, then darken the file afterwards.

Underexpose and brighten, or raise the ISO?

Say the light is fixed: you need 1/125 second to stop movement and f/2.8 is as wide as the lens goes. The meter wants ISO 3200. Shooting at ISO 400 instead and brightening three stops later does not give you a cleaner file. The sensor caught exactly the same number of photons either way, so the shot noise is identical. What differs is the read noise, and that favors the higher ISO, because the camera’s gain lifts the signal above part of the electronic noise before it is added.

On many modern sensors the difference is small, which is called ISO invariance; some have a second, quieter gain mode at a set ISO, called dual native ISO. On older sensors, brightening a dark file shows more noise and sometimes stripes (banding). Either way, the rule holds: the aperture and shutter speed set the noise; ISO mostly sets the brightness. Brightening works well only on raw files. A JPEG has already thrown away much of the shadow data you would need.

Not sure what your ISO setting is really doing if it adds no light? The ISO page works through the gain, the highlight room it costs, and how to choose one from the shutter speed you need.

The Kinds of Noise You Will See

Luminance noise: the grain

Luminance noise is variation in brightness from pixel to pixel. It looks like fine sand over the picture and is the closest thing to film grain. In moderate amounts it reads as texture and many people leave some in.

Test your eye on real photos: pick the one below that was shot at the lower ISO. The grainier frame is the one that caught less light, so look at its smooth areas and shadows first.

Which photo was shot at a lower ISO?

A B

Color noise: the blotches

Color noise (chroma noise) is variation in hue: speckles and soft blotches of magenta, green and blue, worst in dark, flat areas. It almost always looks like a fault. It comes from the color filters over the pixels: each pixel sees only red, green or blue light, so each channel has fewer photons than the whole, and the camera’s color processing mixes their separate noise into false colors.

The blue channel is often the noisiest, especially under warm light such as street lamps or tungsten bulbs. That light holds little blue, so the blue pixels catch the fewest photons, and the white balance then multiplies the blue channel the most to make whites look neutral. Night skies and indoor shadows full of blue speckle are the result.

Fixed pattern noise: hot pixels and banding

Random noise changes from frame to frame. Fixed pattern noise does not: it sits in the same place every time.

• Hot pixels are single pixels that leak more charge than their neighbors and show as bright red, green, blue or white dots, mostly in long exposures.
• Banding is faint horizontal or vertical stripes from the readout electronics. It is usually invisible until you brighten deep shadows several stops, which is one more reason to expose properly rather than rescue a dark file.
• Amp glow is a faint glow along one edge or corner of some sensors in exposures of many seconds, from heat near the readout circuits.

Because it repeats, fixed pattern noise can be measured and subtracted, which random noise cannot.

Heat, Long Exposures and the Dark Frame

A sensor creates a little charge even in total darkness, called dark current, and it grows with time and temperature: roughly doubling every several degrees Celsius of warming. A fraction of a second makes almost none. A two minute exposure on a warm summer night makes plenty, and it shows as hot pixels, a mottled color cast in the shadows, and sometimes amp glow. This is why astrophotographers prize cold nights, rest the camera between long frames, and in dedicated astronomy cameras cool the sensor outright.

Long exposure noise reduction (LENR) is the camera’s fix. After your exposure, it takes a second exposure of the same length with the shutter closed, a dark frame, which records only the heat noise and hot pixels. It then subtracts that pattern from your photo. It works well, but it doubles the wait: a two minute exposure locks the camera for about four minutes. On many cameras the subtraction is applied to the raw file too, so it is not undone by shooting raw. When the wait would cost you the moment, as with a fast moving aurora or a run of frames for star trails, turn it off and shoot a few dark frames yourself at the end, with the lens cap on, at the same settings and temperature, for your software to subtract.

The Noise Ledger: What Each Choice Is Worth

Every way of fighting noise either adds light or averages more photons together. Because both run on the same square root law, they can be compared in one unit, stops. Each line below is computed from that law, for the same scene viewed at the same size.

Three lessons hide in that table. First, the ISO row: turning the dial is not a noise control. Second, sensor size matters because of total light, not pixel size. A bigger sensor at the same f-number and shutter speed is a bigger bucket in the same rain. That is why a full frame camera looks cleaner than an APS-C camera, and both cleaner than a phone, at the same settings (more on this under sensor size). Third, viewing size counts as much as anything you do in the field, which is why comparisons only mean something at the same output size.

Wondering whether a sensor with fewer, bigger pixels is cleaner? At the same print size, mostly not: four small pixels averaged together collect the same light as one big one. Pixel binning is that averaging done in the camera.

Getting More Light Without Getting Blur

The catch is that the two settings that add light both have a price: a wider aperture shrinks the depth of field, and a longer shutter time blurs anything that moves, including your hands. So the practical question is always “where can I find more light that the photo can afford?”

• Open up first when depth of field allows. Going from f/4 to f/2 is two stops, which halves the noise.
• Use a support. A tripod, a wall or a railing lets a still subject take seconds or minutes of exposure at a low ISO. The shutter speed then has to suit only the subject, not your hands.
• Use the slowest shutter the subject allows. People walking need about 1/125, a person sitting still about 1/60. Every halving of the time you can afford is a stop of light lost.
• Add light to the scene. A lamp, a window or a flash on the subject raises the photon count where it matters.
• Stack frames. Several shorter exposures averaged in software collect the light of one long one without its blur or heat. Phone night modes do exactly this, automatically, which is why they look so clean for such small sensors.

Try the support route in the camera simulator: put the camera on a tripod, keep the ISO at 200 or lower, and find the shutter time that gives a well exposed night shot.

And when there is no more light to be had, raise the ISO and take the shot. A sharp photo with some noise beats a clean photo blurred by a slow shutter speed, and a well exposed noisy file beats an underexposed one brightened afterwards. Noise can be reduced later; blur and missed moments cannot.

Noise Reduction and What It Costs

All noise reduction works by guessing which pixel to pixel differences are random and smoothing them, while keeping the differences that are real detail. The guess is never perfect, so every bit of smoothing costs some fine texture: skin pores, grass, fabric, distant leaves. The skill is in deciding how much to give up.

• Color noise reduction is nearly free. The eye sees fine detail mostly in brightness, not in color, so smoothing color speckle costs little. Most raw editors apply some by default.
• Luminance noise reduction costs texture. Pushed too far, it gives the waxy, painted look. Leaving a little fine grain usually looks more natural than removing it all.
• Reduce noise before sharpening. Sharpening raises edge contrast, and grain is full of tiny edges. Sharpen a noisy file first and you sharpen the noise.
• Treat the shadows harder than the subject. A mask that applies more reduction to dark, flat areas, and less to the sharp subject, keeps detail where people look.

In camera, the “high ISO noise reduction” setting changes JPEGs and the preview on the back screen, not the raw data, so a raw shooter loses nothing by setting it low and deciding later. Long exposure noise reduction, as above, is different and often does change the raw file. AI noise reduction tools, trained on pairs of clean and noisy images, now recover far more detail than older methods, especially on raw files processed before demosaicing, though at strong settings they can invent texture that was never there.

Want the step by step version for your editor, sliders and all? The noise reduction guide walks through it, including how to keep detail while you smooth.

When Noise Does Not Matter

Noise you can see at 100 percent on a large monitor is often invisible in the photo people actually look at. A phone screen or a small print shows each image pixel at a fraction of its size, which averages the noise away, as the ledger table shows. Judging every file at 100 percent, called pixel peeping, makes noise seem far worse than it is.

The character of the noise matters as much as the amount. Fine, even grain is easy to live with; coarse blotches are not, even at the same strength. And plenty of photographers want grain. In black and white work especially, a fine luminance grain recalls high speed film grain and gives smooth areas a tooth that pure digital can lack. Photographers such as Daido Moriyama built a style around coarse grain. If you want that look on a clean file, it is better to add grain in editing, where you control its size and strength, than to rely on sensor noise, which is blotchy and colored.

Heading out after dark and wondering which settings to start from? Night photography covers the starting points, and for the sky itself see photographing stars and the Milky Way or the northern lights.

Common Mistakes With Noise

• Blaming the ISO. At a fixed aperture and shutter speed, a lower ISO does not mean less noise; it means a darker file with the same shot noise. Find more light instead.
• Keeping the ISO low “for quality” and brightening later. Same light, so no gain in noise, and on many cameras worse shadows and banding. Use the ISO that makes the picture bright enough.
• Choosing a slow shutter to avoid high ISO when the subject moves. You trade fixable noise for blur that cannot be fixed.
• Underexposing to protect highlights that do not exist. If nothing bright is about to clip, give the sensor more light; check the histogram.
• Judging noise at 100 percent and at a different size from the final use. Look at the size the photo will be seen.
• Turning noise reduction to maximum. Smooth plastic skin and mushy foliage look worse than fine grain. Remove the color noise, then stop well before the texture goes.
• Sharpening before reducing noise, which turns grain into crisp speckle.
• Leaving LENR on for a fast changing sky and missing half the show while the camera takes dark frames.

Try This: Find Where Your Noise Comes From

Ten minutes, a dim room with a lamp, and a camera that shoots raw (or a phone with a manual or pro mode that saves raw). Rest the camera on a table or tripod and frame something with both a lit area and deep shadow, such as a bookshelf beside the lamp. Use the same aperture for all three shots, for example f/4.

1. Shot A, high ISO: set the ISO to 3200 and find the shutter speed that gives a normal exposure, say 1/60.
2. Shot B, same light, low ISO: keep 1/60 and f/4 but set ISO 200. The photo is four stops too dark (200, 400, 800, 1600, 3200).
3. Shot C, more light: keep ISO 200 and lengthen the shutter four stops, from 1/60 to 1/30, 1/15, 1/8, 1/4 second.

In your editor, brighten B by four stops so it matches A and C. Then compare the same shadow area in all three at 100 percent.

How to judge it: C should be clearly the cleanest, with about four times the SNR, because it got sixteen times the light. A and B should look close to each other, because they got the same light; if B is noticeably worse or shows stripes, your sensor is not very ISO invariant, and you have learned to trust the ISO dial over the brightness slider on this camera. Finally, export A at half its width and height and view both copies at 100 percent. The smaller copy looks about as clean as a shot with two stops more light, without touching the camera.

Want to check you can tell what really drives noise? Try the ISO and camera noise quiz.

Noise Questions People Ask

Does a higher ISO cause noise?

Not directly. Noise comes mainly from catching too few photons, which is why you needed the higher ISO in the first place. At the same aperture and shutter speed, a higher ISO usually looks the same or slightly cleaner than a lower one brightened afterwards.

Why are my night photos noisy even at a low ISO?

Probably because they are underexposed and then brightened, either by you or by the camera’s processing. Shadows that caught only a few photons stay noisy whatever the ISO. Give the scene more light with a wider aperture, a longer exposure on a support, or more frames averaged.

Is a camera with fewer megapixels less noisy?

Per pixel, often yes; per photo at the same print size, mostly no. Total light depends on the sensor area, not on how it is divided, and downsizing a high resolution file averages its pixels together. Compare cameras at the same output size.

Should I turn off in-camera noise reduction?

If you shoot raw, the high ISO setting only affects the JPEG and the preview, so it makes no difference to your files. If you shoot JPEG, a low or standard setting keeps more detail. Long exposure noise reduction is worth leaving on unless the extra wait would cost you shots.

Does noise reduction lose detail?

Always some, because the software cannot perfectly tell fine detail from random speckle. Color noise reduction costs very little; luminance noise reduction costs texture as you push it. Newer AI based tools lose much less, but check faces and fine patterns closely at strong settings.

Why do phones handle night scenes so well with such tiny sensors?

Mostly by stacking: a night mode takes many short frames and averages them, which collects far more total light than one exposure, then applies strong noise reduction. A larger sensor still wins when both cameras take a single frame of the same scene.