What Is Image Noise? Where It Comes From and How to Reduce It

You open a frame you shot in a dim room and see the shadows breaking up into grit, with green and magenta blotches crawling across flat walls. We call this noise, and almost everyone blames the same culprit: ISO.

But ISO isn’t the culprit. This whole article is the expansion of a single sentence: noise is not an ISO problem, it’s a light problem. Raising ISO doesn’t create noise; it amplifies a signal that was never collected in sufficient quantity and makes the noise inside it visible. Once that distinction clicks, it becomes obvious which fixes work and which ones are wasted effort.

The what is ISO guide covers ISO’s role in exposure — which value to pick in which situation. This article puts noise itself on the table: where it physically comes from, which setting actually reduces it, and how much of it can be rescued in editing.

The two sources of noise

Noise isn’t one thing. The speckle you see in a frame is the sum of two components with completely different origins, and what you can do about each one differs too.

1. Photon noise — the scene’s own noise

Light isn’t a smooth stream. It arrives in packets called photons. Think of rain: hold a cup out in a shower and the number of drops landing in it each second won’t be constant, it will wobble around an average. Photons behave the same way.

Every pixel on the sensor is a small bucket counting that photon rain. If a bucket receives 100 photons on average, its neighbor gets 92 and the next one 107. That deviation isn’t a defect, it’s the nature of light — and its size is mathematically fixed: the square root of the number of photons collected.

That leads to a very practical conclusion:

• Collect 100 photons and the deviation is around ±10 → signal-to-noise ratio of 10.
• Collect 10,000 photons and the deviation is around ±100 → ratio of 100.

So as light increases, noise grows in absolute terms but shrinks fast in relative terms. That’s exactly why frames shot in plenty of light look clean.

Photon noise has nothing to do with your camera. The most expensive sensor in the world faces the same photon shortage in the same dark room.

2. Read noise — the camera’s own noise

The second component comes from the sensor’s electronics. The circuit that reads the charge in each bucket and converts it to a number adds a small uncertainty every time. This is called read noise.

Read noise is independent of how bright the scene is — it’s always the same amount. That’s why it stays invisible in bright areas next to a large signal, but becomes dominant in deep shadows. When you brighten an underexposed frame in editing, that ugly banded texture appearing in the shadows is usually read noise.

Two smaller items sit alongside these:

• Dark current (thermal noise): The sensor generates charge on its own with heat. It accumulates as exposure time lengthens and the sensor warms up; in multi-minute long exposures it shows up as isolated bright “hot pixels.” It’s more noticeable on a summer night than a winter one.
• Fixed pattern noise: Banding born from tiny manufacturing differences between pixels, identical from frame to frame. Push the shadows far enough and it appears as horizontal or vertical stripes.

What is ISO actually doing?

Now to the root of the common mistake. Raising ISO doesn’t increase the sensor’s sensitivity to light — that sensitivity is fixed. ISO is a gain applied before the charge in the bucket is read out.

The noise floor has a sibling: a sensor’s total dynamic range is set directly by that floor — lower the floor and the range widens at the same time. The maths of that link, and what to do when the scene won’t fit the sensor, are in what is dynamic range.

This has an interesting consequence: because the gain is applied ahead of the readout circuit, raising ISO reduces the relative contribution of read noise. A correctly exposed ISO 3200 frame is usually cleaner than a frame shot in the same light at ISO 400 and pushed three stops in editing.

Most modern sensors are close to ISO invariant: read noise stays nearly constant across the ISO range, so shooting at ISO 1600 and shooting at ISO 400 then pushing two stops give similar results. But “similar” isn’t “identical,” and not every body behaves the same.

There’s also the second base ISO. Many modern cameras switch to a second gain circuit at a particular ISO value (320, 400 or 640 depending on the body), and read noise drops sharply there. Which means a frame shot at ISO 500 can come out cleaner than one shot at ISO 400 — completely counterintuitive, and completely real.

The only thing that truly reduces noise: more light

If the source of noise is a photon shortage, the solution comes down to one heading: get more photons onto the sensor. You only have a handful of ways to do that, and each has a price.

• Open the aperture. Going from f/4 to f/2 quadruples the light — a full two-stop gain in noise terms. The price: shallower depth of field. The relationship between aperture and light is covered in what is aperture.
• Lengthen the exposure. Dropping from 1/125 to 1/30 is again two stops of light. The price: subject or camera movement. The limits of that price are in why photos come out blurry, and the thresholds for handholding are in image stabilization.
• Add light to the scene. This is the most decisive fix and usually the last one people think of. A flash takes you from ISO 6400 to ISO 400 — it ends the noise conversation outright. Flash photography basics covers the first steps; ring flash in macro is the clearest example of added light erasing noise.
• A bigger sensor. Shooting the same framing at the same f-number, a larger sensor collects more total light thanks to its wider physical aperture. That — not “a better sensor” — is the real reason full frame looks cleaner than APS-C in low light. The full comparison is in full frame vs APS-C.
• Shoot many frames and stack them. Averaging n frames of the same scene reduces noise by √n: 4 frames buy one stop, 16 frames buy two. It’s standard practice in astrophotography, and it’s usually the first advanced step on top of the settings in Milky Way settings.

Expose to the right (ETTR): free cleanliness

Another way to get the most photons onto the sensor is to use the brightest exposure you can without clipping. This is called exposing to the right, named for the histogram peak leaning toward the right side.

The logic is simple: in a digital exposure, most tonal information is stored in the upper stops. Push the frame as far right as you safely can and the shadows collect more photons, which means they stay surprisingly clean when you pull them back down in editing.

Exposing to the right only pays off if you shoot raw: in a JPEG that extra tonal information has already been thrown away. The difference is detailed in raw vs JPEG. To see how many stops your camera’s meter is off, metering modes and exposure compensation will make your life easier.

Color noise or luminance noise?

The speckle in a frame has two different faces, and our eyes react to them very differently:

• Color noise (chroma): Green, magenta and red blotches on flat surfaces. Nothing in nature looks like this, so the eye instantly reads it as “broken.” The good news: it’s easy to clean and costs almost no detail.
• Luminance noise: A grainy, sandy texture. Because it resembles film grain, the eye is far more forgiving of it — sometimes it even likes it. The bad news: cleaning it eats detail directly.

Cleaning noise in editing

In classic noise reduction the order isn’t negotiable, and it slots into the chain described in the raw editing workflow:

There’s also a local approach: noise bothers us mostly in flat areas (sky, walls, shadows) and is nearly invisible in detailed ones. Using masking tools to apply noise reduction only to the sky always beats softening the entire frame.

AI-based denoisers

The biggest leap of recent years lives here. Lightroom’s Denoise, DxO’s DeepPRIME and their peers work on raw data in its most untouched state — before the color information is interpreted. That lets them make a distinction the classic sliders can’t reach: they’re far better at guessing what is texture and what is noise.

In practice the gain is around two stops: an ISO 12800 frame starts to look like ISO 3200. But three things are worth knowing:

• The order changes. These tools run at the very start of the raw process and usually produce a new DNG; the classic “color → luminance → sharpening” sequence applies to the steps after that.
• The cost is disk space and time. The resulting DNG can be several times larger than the original, and processing takes seconds per frame depending on your GPU.
• At the extremes it starts inventing. On very noisy frames the algorithm can hallucinate detail that was never there: strands of hair flatten out, distant text looks legible but reads wrong. Check the result at 100% zoom.

When noise isn’t a problem

Before fighting noise, ask the real question: where is this frame going to be seen?

• Noise disappears as size shrinks. Downsizing a 6000-pixel-wide frame to 1500 pixels averages neighboring pixels and cuts noise by roughly a stop. Spending hours at 100% zoom cleaning noise nobody will see on Instagram or a web page is wasted work.
• Paper forgives. On matte paper, noise shows far less than it does on screen.
• Black and white loves noise. Color noise disappears by definition, and what’s left turns into something like film grain. Black and white photography explains why that texture is often preserved on purpose.
• In contrasty, dark scenes the eye isn’t looking for it. In night street photography, a gritty texture usually adds atmosphere.

The decision order in the field

Looking through the frame in a dim room, make the noise decision in this order:

How to build the whole set of settings in low light is covered step by step in shooting in low light without flash.

Common mistakes

• Blaming ISO for noise: Lowering ISO and darkening the frame increases noise, it doesn’t reduce it. The culprit is the shortage of light.
• Planning to shoot dark and brighten later: Read noise and fixed pattern noise are waiting for you in the shadows. Expose correctly at capture.
• Never measuring your camera’s ISO ceiling: Most photographers stay stops below their body’s real limit. Run the test above once and you’ll never hesitate again.
• Over-cleaning luminance noise: Waxy skin and plastic leaves stand out far more than grain does.
• Sharpening before reducing noise: Sharpened noise is permanent.
• Judging noise at 100% zoom: Nobody views the photo at that size. Judge it at the size you’ll publish it.