How Photography Works | Dithering and Posterisation — Why Noise Can Make Quantisation Look Smoother

A sunset can be physically smooth and photographically stepped.

The sky may change continuously from deep blue to pale orange. But a digital representation has a finite number of numerical levels. If too few levels survive—or if later editing stretches a narrow range too far—the gradient can break into visible bands.

Bit depth determines how many discrete numerical levels are available to represent continuous-looking tonal or colour variation.

This article continues the canonical How Photography Works | Every Photograph Leaves Something Out knowledge map. Tone Curves explains how brightness relationships are remapped. Colour Spaces and Gamut explains the coordinate range. Bit depth asks how finely values inside that range can be represented.

Quick Read

Bit depth sets the number of discrete code values available per channel or sample. An 8-bit channel provides 256 possible code values; higher bit depths provide more. More levels reduce the numerical spacing between adjacent values and give processing more room before quantisation steps become visible. Banding occurs when a smooth underlying gradient is represented by too few distinguishable output levels, especially after strong tonal or colour edits. Noise or deliberate dithering can sometimes hide the boundaries by breaking up coherent steps, but dithering does not create new underlying scene precision. File format, colour space, transfer curve, processing sequence and final display all influence whether banding becomes visible.

continuous scene → sampled values → finite code levels → processing → display → smooth gradient or visible bands

Continuous Light Meets Discrete Numbers

The physical scene can vary smoothly. A digital file cannot store an infinite continuum. It assigns measurements to finite numerical codes.

If many nearby physical values are mapped to one code, they become indistinguishable in that representation. This is quantisation.

Eight Bits Means 256 Code Values Per Channel

An 8-bit channel can represent values from 0 to 255: 256 discrete codes. In a conventional three-channel RGB image, combinations of those channel values permit millions of colour triplets.

That does not mean every triplet is perceptually distinct or that all code values are spaced uniformly in physical luminance. Transfer functions and colour spaces determine the visual meaning of those numbers.

Higher Bit Depth Gives Finer Numerical Steps

A 10-bit channel has 1024 codes. A 12-bit channel has 4096. A 14-bit channel has 16,384. A 16-bit integer channel can represent 65,536 levels.

The practical benefit is not that viewers can individually name all those levels. It is that transformations can redistribute and compress values while leaving enough intermediate distinctions to avoid visible stepping.

Camera RAW Bit Depth Is Not the Same as Final JPEG Bit Depth

A camera may digitise sensor measurements at 12, 14 or another bit depth, while a conventional JPEG export stores 8 bits per channel after demosaicing, white balance, colour conversion and tone mapping.

The earlier high-precision values provide processing headroom. The final lower-bit-depth file can still look excellent because the values have already been mapped into a display-oriented range.

More Bits Do Not Create More Photons

If a sensor measurement is dominated by noise, storing it with more numerical precision does not make the underlying signal more certain. Bit depth describes coding precision; Noise describes measurement uncertainty.

There is little benefit in representing random uncertainty with exquisite numerical precision if the physical signal itself cannot support those distinctions.

Banding Appears When Adjacent Codes Become Visible

Imagine a smooth blue sky represented by only ten brightness levels. Instead of continuous change, you would see ten broad stripes. Real 8-bit images have far more levels, but strong editing can effectively spread a narrow original range across a large output range and expose the steps.

Banding is therefore a visibility problem created by quantisation plus rendering conditions.

Smooth Gradients Are the Best Place to See It

Clear skies, studio backdrops, fog, walls and out-of-focus gradients contain little texture to hide tonal steps. The visual system can follow the boundary between adjacent code levels across a large area.

Busy foliage may contain the same numerical limitations without obvious bands because texture breaks up the contours.

Strong Curves Can Turn Hidden Quantisation Into Visible Steps

Suppose a gentle shadow gradient occupies only a small portion of the available code range. Apply a steep tone curve and those few input levels are spread over a much larger output interval. Gaps between the values become easier to see.

This is why heavy editing is safer in higher-precision working data. Tone Curves can redistribute contrast only from distinctions that still exist.

White-Balance Corrections Can Stress One Channel

Large channel multipliers used to correct unusual illumination can spread weak channel values apart and expose noise or quantisation. A strongly under-recorded blue channel under warm light, for example, may need substantial amplification.

This connects to White Balance: the colour correction can only work with the precision and signal that capture preserved.

Wide Gamut and Bit Depth Interact

A wide colour space spreads representable colours across a larger coordinate region. At the same fixed bit depth, neighbouring codes can correspond to larger colour differences in some parts of the space.

This does not make wide gamut bad. It means high-precision editing and wide-gamut workflows belong naturally together when aggressive processing is expected.

Transfer Curves Redistribute Code Values

Display-oriented spaces do not usually allocate code values linearly with physical light. Nonlinear transfer curves devote numerical precision in a way better matched to perception and practical displays.

That is one reason 8-bit display images can appear smooth under normal conditions despite containing only 256 codes per channel.

Dithering Can Hide Band Boundaries

Add a small amount of carefully distributed noise before reducing precision and neighbouring pixels may alternate among available levels rather than forming one coherent contour. At normal viewing distance, the eye integrates the pattern and perceives a smoother gradient.

This is dithering. It replaces structured quantisation error with less objectionable spatial variation.

Dithering Does Not Restore the Original Continuum

The file still contains finite code values. Dithering distributes those values across space so average local appearance can approximate intermediate tones.

The representation becomes perceptually smoother without increasing the fundamental number of available codes.

Noise Can Accidentally Act Like Dither

A perfectly smooth synthetic gradient can band more visibly than a photograph containing fine sensor noise. The random variation breaks up the coherent boundary between levels.

This is one reason aggressive denoising can sometimes reveal banding that was previously hidden by natural noise.

JPEG Compression Can Add Its Own Quantisation

Even if an image begins with adequate bit depth, lossy JPEG encoding quantises transform coefficients. Smooth gradients can therefore acquire banding, blocking or ringing after strong compression.

Article 34, JPEG Compression, owns that additional lossy stage.

Resizing Can Hide or Reveal Banding

Downsampling may average neighbouring values and make bands less visible. Enlargement can make existing bands wider and easier to notice. Sharpening or local contrast can emphasise their edges.

The final appearance therefore depends on the entire output pipeline, not only the master file’s nominal bit depth.

Display Bit Depth Sets Another Boundary

A high-precision image shown through a limited display pipeline may still band. Conversely, temporal or spatial dithering in a display can simulate intermediate levels and make gradients appear smoother.

The file may be 16-bit while the visible chain contains an 8-bit bottleneck. The weakest stage can become the visible limit.

Printing Has Different Quantisation Behaviour

Printers create tone through patterns of ink or pigment dots, screening, droplet sizes and paper interaction. The mechanism is not a direct one-code-value-to-one-visible-level mapping like a simple display model.

Nevertheless, the source file still needs enough precision and a suitable conversion pipeline to avoid introducing posterisation before the printer performs its own halftoning or dithering.

Posterisation Is Banding Used or Exposed More Broadly

Posterisation describes the reduction of continuous tonal variation into a small number of visibly distinct regions. It can be an artefact of limited precision or a deliberate creative effect.

The physical operation is similar; the photographic job changes whether the result is failure or style.

Higher Precision Helps Most During Manipulation

A well-prepared 8-bit delivery file can look excellent. The advantage of 16-bit or other high-precision workflows is especially strong while large curves, colour moves, compositing and repeated transforms are still being applied.

More intermediate levels reduce rounding accumulation and preserve small distinctions until the final output mapping is settled.

Higher Bit Depth Does Not Guarantee Better Final Colour

A 16-bit file with the wrong profile, clipped channels or poor source data can be less faithful than a correctly managed 8-bit file. Numerical precision is one layer, not the whole photography system.

Colour Spaces and Gamut supplies the coordinate meaning; bit depth supplies the coordinate granularity.

Scientific Imaging Needs the Original Precision Path

If numerical pixel differences are being analysed quantitatively, reducing bit depth or applying display-oriented transformations can alter measurable values. A presentation image may legitimately be 8-bit while the underlying analysis uses higher-precision calibrated data.

Scientific Photography Is Measurement With a Camera keeps source measurement and display rendering separate.

Banding Can Be Evidence About Processing, Not the Scene

A clear sky with hard tonal rings may reveal low-bit-depth processing, aggressive curves, compression or platform conversion rather than atmospheric layers. The artefact belongs to the representation path.

This is another reason Versions matters. A derivative can acquire structures absent from the original photograph.

The Bit-Depth Audit

1. Source precision: what bit depth or numerical format did the source use?
2. Working precision: is editing occurring at higher depth?
3. Gradient: are large smooth tonal regions present?
4. Curves: have narrow ranges been stretched aggressively?
5. Colour space: how widely are code values distributed?
6. Denoising: has natural dither-like noise been removed?
7. Compression: has lossy quantisation added another limit?
8. Display: does the output chain support the file’s precision?
9. Dither: should structured error be decorrelated before reduction?
10. Purpose: editing master, delivery image, creative posterisation or measurement?

Photography Laboratory 1: Build a Gradient

Create or photograph a smooth gradient. Make versions at different processing depths if your software supports them, then apply a strong curve. Compare where bands become visible.

Photography Laboratory 2: Noise as Dither

On a copy of a banded gradient, add a very small amount of fine noise. View at normal size. Observe whether coherent bands become less visible even though the image now contains more pixel variation.

Photography Laboratory 3: Compression Stress

Export a smooth sky gradient at high and low JPEG quality. Compare banding, blocks and edge artefacts. Keep the high-precision source untouched.

For Primary Readers

Draw a smooth rainbow with many crayons, then try again with only four colours. The fewer available steps make the boundaries easier to see.

For Secondary Readers

Connect bit depth to powers of two and quantisation. Explain why more code values make the numerical gaps between representable levels smaller.

For Advanced Readers

Analyse quantisation error, transfer-function allocation, integer versus floating-point processing, dithering and error decorrelation. Connect visible banding to just-noticeable differences, output luminance, gamut volume and accumulated rounding across nonlinear transforms.

Common Misconceptions

• “16-bit means sixteen million times better than 8-bit.” More codes provide processing precision; visible benefit depends on the entire workflow.
• “More bits reduce sensor noise.” Coding precision and measurement uncertainty are different.
• “Banding always came from the sky.” It can be introduced by processing or display.
• “Dithering creates real new tones.” It spatially distributes existing discrete levels to make transitions look smoother.
• “8-bit images are always poor quality.” Well-rendered 8-bit delivery images can look excellent under normal conditions.

Frequently Asked Questions

Why does banding appear after editing?

Strong edits can spread a small set of neighbouring input values across a larger output range, making quantisation gaps visible.

Does 16-bit eliminate banding?

It greatly increases numerical precision during editing, but later conversion, compression, display limits or extreme transforms can still create banding.

Why can adding noise reduce visible banding?

Fine random variation breaks up coherent quantisation boundaries so the eye integrates neighbouring values into a smoother-looking transition.

Final Thought: Smoothness Requires Enough Room Between the Numbers

A continuous-looking photograph is built from discrete values. It succeeds when those steps are fine enough, distributed well enough and rendered gently enough that the viewer experiences the gradient rather than the staircase underneath.

Bit depth is not how much beauty the photograph contains. It is how finely the representation can divide the path between one value and the next.