Is airlight always white?

No. Its color depends on the illumination and atmosphere. Models often represent A as a three-channel color value.

Why do distant objects often look more washed out?

A longer path through a scattering atmosphere can reduce transmission and increase the relative airlight contribution under common conditions.

Can one hazy photograph determine the exact clear scene?

Generally not without additional assumptions or information. Different combinations of scene radiance and atmospheric parameters can explain an observation.

AI Glossary / Computer vision and multimodal AI / Intermediate

Airlight

Definition

Airlight is light scattered by the atmosphere into the camera's line of sight. In a common haze model, it adds a veil-like contribution to the image, reducing the contrast of distant objects and shifting their appearance towards the atmospheric light color.

What it means

A camera looking through haze receives more than light traveling directly from an object. Part of the object's light is scattered away, while environmental illumination is scattered towards the camera along the viewing path.

These are related but distinct effects: attenuation reduces the direct scene contribution, and airlight adds a scattered contribution. Together they can make a dark distant object look brighter and less distinct. Airlight need not be perfectly white; its color depends on illumination and atmospheric conditions.

Understanding this term helps explain why ordinary brightness adjustment is not a complete physical description of dehazing. The apparent washout can vary across a scene as viewing distance and atmospheric transmission change.

How it works

A widely used simplified model is I(x) = J(x)t(x) + A(1 − t(x)). I is the observed image, J is the desired scene radiance, t is transmission, and A represents atmospheric light, often approximated as constant across the image. The term A(1 − t) is the airlight contribution.

When transmission is near one, most of the direct scene contribution remains and the airlight term is small. When transmission is low, the observation moves towards A and contains less direct information about J. Color images apply the relationship across channels.

Dehazing methods estimate transmission and atmospheric light using assumptions, additional measurements or learned models. A single observed image does not uniquely reveal every unknown. Bright surfaces, sky, uneven illumination and complex weather can challenge simplified assumptions, so an estimated correction should not be treated as an exact physical recovery.

Three practical examples

A landscape-editing tool estimates atmospheric effects to make distant ridges appear less washed out.
A vision researcher generates synthetic hazy examples by combining clear images with assumed transmission and atmospheric-light values.
An outdoor-camera team examines whether a dehazing stage helps object detection under representative visibility conditions.

Think of it this way

Think of viewing a sign through a faintly glowing curtain. The curtain both weakens the sign's contribution and adds its own brightness to the view. The analogy captures two effects, but a real atmosphere scatters light throughout a volume; it is not a single sheet, and the simple model may omit important directional and multiple-scattering behavior.

Worked example

Consider one normalized brightness channel with clear-scene value J = 0.2, transmission t = 0.4 and atmospheric light A = 0.9. The direct contribution is 0.2 × 0.4 = 0.08. Airlight contributes 0.9 × 0.6 = 0.54. The observed value is therefore I = 0.62.

Under these assumed values, a dark object appears much brighter because the scattered contribution dominates. If A and t are known, rearranging gives J = (I − A)/t + A = (0.62 − 0.9)/0.4 + 0.9 = 0.2.

In practice they are estimates. Dividing by a very small transmission amplifies errors and noise, so restoration methods commonly restrain that division. The arithmetic is an illustration of the model, not evidence that a real scene exactly follows it.

Illustrative case study

Fictional example: Ridgewatch Media records scenic footage from a fixed viewpoint. On hazy days, distant hills become pale while nearby trees remain comparatively contrasty.

The editor compares a dehazing tool with a simple global contrast adjustment. They inspect sky gradients, bright buildings and fine branches, where an aggressive correction may create unnatural colors or dark outlines. The original footage is retained, and the processed version is labeled as an edited presentation.

The team evaluates continuity between frames as well as single-image appearance. This fictional case shows why varying atmospheric effects need careful review; it does not claim that the edited scene reveals details absent from the capture.

Common misconceptions and mistakes

Airlight is not simply a camera exposure error, and it is not identical to lens flare. Different physical mechanisms may produce superficially similar veils.

The atmospheric-light parameter A should not be confused with the full airlight contribution A(1 − t), which varies with transmission. Nor is the brightest image pixel necessarily a reliable estimate of A; it may be a white object or a lamp. Removing too much estimated airlight can distort colors and exaggerate noise. The common assumption that A is spatially constant is a modeling convenience, not a universal atmospheric law.

Business use

Airlight modeling can inform enhancement, simulation and robustness testing for outdoor imagery. Its business value depends on whether the resulting images or decisions improve for the intended task.

For presentation, compare visual plausibility and temporal consistency. For automated analysis, measure detection or measurement errors using suitable reference data. Keep the original image available and record processing settings. A visually stronger image is not automatically a more faithful measurement, and a model developed for daytime haze may need separate evaluation under other lighting or scattering conditions.

Check your understanding

In I = Jt + A(1 − t), which term represents airlight, and what happens to it as t approaches one?

Answer: A(1 − t) represents airlight. It approaches zero as transmission approaches one.

Related terms

Dark channel prior
Contrast
Atmospheric scattering
dehazing
transmission map
scene radiance
image restoration
haze model

Learn next

Use the worked model to calculate observed brightness for transmission values of 0.2, 0.6 and 0.9 while holding J and A fixed. Observe how the result moves towards the atmospheric-light value as transmission falls. Then study how a dehazing method estimates its unknowns and identify which scene types violate its assumptions.

Sources

He, Sun and Tang, Single Image Haze Removal Using Dark Channel Prior, original research paper: https://people.csail.mit.edu/kaiming/publications/pami10dehaze.pdf Author project page: https://people.csail.mit.edu/kaiming/cvpr09/index.html The numerical values and Ridgewatch case are original illustrations. The article describes a simplified image-formation model, not a full atmospheric simulation.

Previous: Air gap (networking) Next: AIXI