AI Glossary / Computer vision and multimodal AI / Intermediate

Tone mapping

Definition

Tone mapping transforms image brightness values to suit a display or output medium, especially when the captured or rendered scene has a wider dynamic range than the output can reproduce. It balances overall brightness, contrast, and visible detail. The result is a rendering choice, not a direct preservation of every original luminance relationship.

What it means

A scene can contain bright sunlight, deep shadows, and intermediate surfaces at the same time. A screen or print may not reproduce their full brightness range. Simply clipping bright values or scaling everything down can sacrifice useful detail.

Tone mapping chooses how to fit that information into the available range. A global operator applies one mapping to a given input value everywhere. A local operator also considers surrounding content, allowing different treatment of regions with different illumination or structure.

The process is separate from acquiring high-dynamic-range information. Combining exposures or recording a wide-range image supplies data; tone mapping determines how that data is presented. If capture already lost a highlight, remapping its clipped value does not recover the missing original detail.

How it works

A pipeline interprets the input in an appropriate working representation, selects an exposure or adaptation scale, and applies a brightness transformation. Color handling and output encoding are also needed so the result is displayed as intended.

A simple global compression curve maps increasing brightness to increasing output values while progressively reducing the separation of very bright inputs. Local methods can separate broad illumination variation from smaller detail and compress those components differently. Durand and Dorsey's bilateral approach is one example of that strategy.

Choices involve tradeoffs. Strong compression can make a scene look flat; aggressive local processing can produce halos or exaggerate texture. For video, inconsistent choices between frames can create flicker. The desired appearance and target viewing conditions help determine which compromises are acceptable.

Three practical examples

1. An architectural photographer prepares an interior image containing bright windows. Tone mapping can make both the room and outside view visible when the source captures them, while preserving a believable sense of illumination.

2. A renderer converts a high-range lighting simulation into an image for a standard display. The chosen operator affects how highlights and shadow details appear, so comparisons should use consistent rendering settings.

3. A media team creates outputs for displays with different capabilities. It reviews each target rendering rather than assuming one set of pixel values will look equivalent on every device.

Think of it this way

Think of fitting a long range of elevations onto the limited height of a relief model. You must compress some differences while keeping useful features recognizable. The analogy explains range reduction. It is limited because visual appearance depends on local contrast, color, adaptation, and display conditions, not only on preserving an ordering of numbers.

Worked example

Take a simplified global curve y = x/(1 + x), where x is a nonnegative brightness value already scaled into the operator's working units. This is the basic compressive form associated with the global photographic operator, not its complete exposure-selection pipeline.

For x = 0.1, the output is approximately 0.091. For x = 1, it is 0.5. For x = 10, it is approximately 0.909. For x = 100, it is approximately 0.990.

The input range from 0.1 to 100 spans a brightness ratio of 1,000 to one. The corresponding positive output values span a ratio of about 10.89 to one. All outputs remain below one, so high inputs approach the upper limit smoothly rather than being abruptly clipped by this formula.

However, inputs 10 and 100 are now separated by only about 0.081. Compressing range also compresses distinctions. The example demonstrates the mapping before output encoding and does not specify an actual display's luminance or guarantee perceptually equal steps.

Illustrative case study

Fictional case study: RoomVista prepares property images from wide-range captures. An initial local tone-mapping preset reveals extensive shadow detail but creates bright halos around dark window frames and makes every room look equally illuminated.

The editor compares a milder local treatment with a global mapping and reviews the image at its intended viewing size. The team keeps the captured source and records the approved output settings. It checks whether the presentation remains consistent with the photographed space rather than maximizing visible detail at every location.

Different rooms receive different adjustments where necessary. This fictional case illustrates artifact review and presentation choices, not a claim that a particular operator reproduces exactly what every viewer would have perceived in the room.

Common misconceptions and mistakes

Do not use HDR capture and tone mapping as interchangeable terms. One concerns available scene information; the other concerns its rendering for an output.

Do not assume that preserving more local detail always produces a more faithful appearance. Flattened illumination or exaggerated edges can make a result less believable.

Do not treat a mathematically valid curve as a complete color pipeline. The input scale, working color representation, output transfer function, and display behavior affect the final result. Applying the same mapping at the wrong stage can produce unexpected brightness or color.

Business use

Tone mapping matters in photography, rendering, video, and visual communication whenever source and output ranges differ. It can improve readability of detail, but it also changes the visual emphasis of a scene.

Define target outputs and review criteria before choosing an operator. Include difficult highlights, deep shadows, and strong boundaries in comparisons. For video, inspect transitions over time. Retain high-range masters when available so future display versions can be rendered from the source rather than from an already compressed tonal rendition.

People also ask

Is tone mapping always local?

No. Global methods use a single value mapping, while local methods also depend on spatial context.

Can it recover clipped highlights?

Not reliably from the clipped values alone. Those distinctions must be present in the source or supplied by additional evidence.

Does it always produce the exaggerated HDR look?

No. That is one possible processing style. Many tone-mapped images aim for restrained, photographic appearance.

Check your understanding

A curve maps very large brightness values close to one. Does fitting them within range mean their original differences are fully preserved?

Answer: No. They may be compressed into a narrow output interval, reducing their separability even without hard clipping.

Related terms

• Dynamic range
• Tone curve
• exposure
• local contrast
• bilateral filter
• display encoding

Learn next

Compare a global compression curve with a local operator on the same wide-range image. Inspect shadow separation, highlight roll-off, edge halos, and color before deciding which rendering fits the intended output.

More in Computer vision and multimodal AI

• Texture - Texture is the visual pattern formed by variations in an image region, such as the grain of wood, a fabric weave, or a speckled surface.
• Texture synthesis - Texture synthesis creates new image content with the visual character of a texture, such as woven fabric, stone, or scattered leaves.
• The Bitter Lesson - The Bitter Lesson is Richard Sutton's 2019 argument that broadly applicable methods able to use increasing computation, especially search and learning, have...
• Train/validation/test split - A train/validation/test split divides a dataset into parts with different roles: training fits a model, validation guides development choices, and testing...
• Transcript-based editing - Transcript-based editing is a way to edit recorded speech and its associated media by selecting and arranging words in a time-linked transcript.
• Trifocal tensor - The trifocal tensor is a mathematical representation of the projective geometry connecting three camera views.

Sources

Erik Reinhard and colleagues, Photographic Tone Reproduction for Digital Images, research page and paper: https://w3.impa.br/~lvelho/ip02/papers/reinhard/

Frédo Durand and Julie Dorsey, Fast Bilateral Filtering for the Display of High-Dynamic-Range Images: https://people.csail.mit.edu/fredo/PUBLI/Siggraph2002/

The numerical inputs and fictional property-image review are original teaching illustrations.

Previous: Tone curve Next: Top-down design