Can the same input brightness map to two output values?

Yes. Different neighborhoods can lead a local operator to treat equal input values differently.

Does it recover clipped highlights?

No. Tonal remapping cannot determine detail already lost through clipping.

Why can halos appear?

Unsuitable neighborhood or scale processing can create artificial brightness transitions near strong boundaries.

AI Glossary / Computer vision and multimodal AI / Intermediate

Local tone mapping

Definition

Local tone mapping adjusts an image's tonal values using information about each location's surroundings. It often compresses a wide brightness range for display while trying to retain useful local contrast. Unlike a single global curve, a local operator can map equal input values differently when they occur in different neighborhoods.

What it means

A bright outdoor view and a dim interior may both contain important detail, yet one display may not reproduce the full captured brightness range. A global adjustment applies the same value-to-value mapping everywhere. A local method can respond to nearby structure and brightness, allowing different regions to receive different treatment.

This is a family of techniques rather than one algorithm. Methods differ in their neighborhood analysis, scale decomposition, edge handling, color treatment, and intended visual result.

The output is a rendering for viewing. It does not necessarily preserve the original ratios of scene luminance, and it should not automatically be used as quantitative brightness data. Making details more visible is different from reconstructing missing information or measuring the scene more accurately.

How it works

Determine a local adaptation value, a multiscale representation, or another neighborhood-dependent description. Use it to guide how brightness and contrast are transformed at each location.

One influential approach decomposes log intensity into a broad base layer and a detail layer using an edge-preserving filter. It compresses the base layer's range while retaining much of the detail, then recombines them and handles color for display.

Other methods use different constructions, including local Laplacian processing. Their shared feature is spatially varying tonal treatment, not a mandatory particular filter.

Check boundaries carefully. If neighborhood estimation blends across a strong edge inappropriately, the result can show bright or dark halos. Video introduces an additional requirement: local adjustments should remain stable over time instead of flickering as scene content changes.

Three practical examples

1. Interior presentation: An editor prepares a high-contrast room image so viewers can see relevant details in both shaded furnishings and brighter areas.

2. Render review: A design team maps a high-dynamic-range rendering onto a conventional display to inspect local details while documenting the viewing transform.

3. Camera processing research: Engineers compare local and global operators on controlled scenes, evaluating visibility, artifacts, color, and consistency across frames.

Think of it this way

Think of a skilled printer adjusting exposure in different parts of a photographic print so useful details remain visible. The analogy captures local control, but an automatic operator follows mathematical neighborhood rules. It does not necessarily know which details tell the story or whether its result matches the original scene's appearance.

Worked example

Consider a simplified decomposition in base-two logarithmic intensity. At one point, the log intensity is 6 and the estimated base is 5.5, leaving detail of 0.5. At a second point, log intensity is 10 and base is 9.5, again leaving detail of 0.5.

Compress the base by multiplying it by 0.5 while keeping the detail unchanged. The first output log value becomes 2.75 + 0.5 = 3.25. The second becomes 4.75 + 0.5 = 5.25.

The original log difference was 4, corresponding to an intensity ratio of 16. The new difference is 2, corresponding to a ratio of 4. In this toy calculation, the broad contrast is reduced while the chosen detail contribution is retained.

A complete tone mapper must also establish output brightness, handle color and display limits, and choose an appropriate decomposition. The calculation is a teaching illustration of base-detail processing, not the exact implementation of a named operator. It does not guarantee that every local edge or texture survives unchanged.

Illustrative case study

Fictional case study: An architecture practice prepares a visualization with sunlit windows and a dark reception area. A strong local tone-mapping preset reveals furniture texture but creates pale bands around window frames.

The visualization team compares a milder setting, a different edge-preserving decomposition, and a global reference rendering. It reviews the images at the intended presentation size and checks that material colors remain credible.

For an animation, it also checks consecutive frames because individually attractive adjustments can fluctuate over time. The final presentation uses a documented viewing workflow. The case illustrates balancing visibility and artifacts, without assuming that the most detailed-looking result is the most faithful representation.

Common misconceptions and mistakes

Local tone mapping is not simply increasing sharpness. It changes tonal relationships using spatial context, although some settings can also emphasize detail.

An HDR-looking result does not prove that the source captured high-dynamic-range information. Aggressive processing of a limited image can amplify noise or reveal clipping without restoring lost detail.

Equal input values need not produce equal output values under a local operator. That can be intentional, but it changes how the output should be interpreted.

Halo-free behavior on one test image does not guarantee artifact-free results everywhere. Neighborhood scale and edge handling should be evaluated across representative content.

Local processing is not universally superior to a global curve. The appropriate choice depends on the scene, desired appearance, display, and consistency requirements.

Business use

Local tone mapping can make demanding imagery more useful for presentation and visual inspection when a simple global transform hides important structure. It also introduces additional processing choices that affect appearance and reproducibility.

Define the intended display and visual objective. Evaluate local contrast, halos, noise visibility, color shifts, and temporal stability where relevant. Preserve the original high-range data and record the viewing transform. If later analysis requires luminance relationships, use the appropriate source representation rather than inferring them from the locally adjusted display image.

Check your understanding

A local operator changes a broad intensity ratio from 32 to 8. Has it preserved that original ratio?

Answer: No. It has compressed the ratio, even if nearby texture remains visible. The display result should not be treated as unchanged scene-luminance data.

Related terms

Tone mapping
Dynamic range
Local Laplacian filters
local contrast
bilateral filter
local Laplacian filter
halo artifact
display transform

Learn next

Compare global and local mappings on an image containing both strong edges and low-contrast texture. Trace a base-detail decomposition and inspect what happens when the neighborhood size or compression strength changes.

Sources

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

The primary paper was opened and its spatially varying and base-detail explanation checked. Log-intensity arithmetic and the architecture scenario are original teaching material.

Previous: Local tangent space alignment Next: Locale (computer hardware)