Checker–Shadow

The Checker–Shadow (Adelson) illusion — a clear, layered explainer

What you see

In Adelson’s Checker–Shadow illusion, two marked squares on a chequerboard (A and B) look very different in shade—A appears dark, B looks light—yet they’re physically identical (same pixel values / print ink). A sits in the light; B sits in the shadow of a cast object. Remove or “reveal” the context and the equality becomes obvious. First published by Edward H. Adelson in 1995, it’s now the go-to demonstration of how strongly context and inferred illumination shape perceived lightness (Wikipedia, n.d.; MIT Perceptual Science Group, n.d.).

How to make it pop (and how to kill it)

Keep a high board contrast: a light–dark chequerboard with a soft-edged cast shadow across it. The soft edge reads as illumination, not paint (MIT Perceptual Science Group, n.d.).
Choose the labelled tiles so that one is a light square in shadow and the other is a dark square in light; surround each with oppositely shaded neighbours to amplify simultaneous-contrast cues (MIT Perceptual Science Group, n.d.).
The illusion fades if you harden the shadow edge (it starts to look like paint), reduce global contrast, or remove the 3-D cues to shadow and occlusion (MIT Perceptual Science Group, n.d.).

A short history (and why it stuck)

Adelson developed the image to highlight that the visual system aims for lightness constancy—estimating surface reflectance despite changes in illumination—and that mid-level organisation (grouping, junctions, shadows) can dominate over simple local contrast. His 1993 Science paper laid the groundwork; the 1995 figure became iconic, and a widely used classroom talking point (Adelson, 1993; MIT Perceptual Science Group, n.d.).

The classical ingredients

Simultaneous contrast: each tile’s apparent lightness is nudged by the local surround (dark neighbours make a tile look lighter, light neighbours make it look darker). This on its own is not enough for the full effect, but it contributes (Adelson, 1993).
Illumination discounting: the brain treats the soft gradient across the shadow as a change in lighting, not paint, and “divides out” that lighting to recover surface reflectance. As a result, a light square in shadow is interpreted as high-reflectance despite its low luminance, so it looks light; a dark square in light goes the other way (Adelson, 2000; MIT Perceptual Science Group, n.d.).
Perceptual organisation: cues such as T-junctions / occlusion, shadow geometry, and soft edges make a 2-D picture behave like a 3-D scene in your head. That mid-level parsing is central to the illusion’s strength (Anderson, 1997; Albert, 2007).

Where “simple” accounts fall short

Purely retinal stories (e.g., lateral inhibition) explain simultaneous contrast, but they can’t justify why soft shadows, junction structure, and scene interpretation so powerfully change the percept. Adelson’s demonstrations, and later work on transparency/occlusion, show that perceptual organisation and layered scene decomposition are needed alongside early spatial filtering (Adelson, 1993; Anderson, 1997).

The contemporary view: combining mechanisms

Most researchers now treat Checker–Shadow as a multi-stage computation:

Early spatial filtering & adaptation provide local contrast signals (think Retinex-style normalisation) (Land & McCann, 1971; Jameson & Hurvich, 1961).
Mid-level organisation segments the image into surfaces and illumination/shadow fields using junctions, soft edges, and 3-D cues, then assigns lightness by discounting the inferred illumination (Anderson, 1997; Knill & Kersten, 1991).
Anchoring frameworks (Gilchrist) formalise how regions set reference levels for lightness within grouped contexts, accounting for many “errors” including Checker–Shadow (Gilchrist, 2006).

No single mechanism nails every lightness illusion, but illumination-aware, organised models explain why B “must be” light (it’s in shadow) and A “must be” dark (it’s in light), even when their luminance is identical (Adelson, 2000).

Parameter sensitivities (practical notes)

Shadow edge: soft = strong illusion; hard = looks like paint → weaker effect (MIT Perceptual Science Group, n.d.).
Neighbour context: maximise by choosing A and B so their local surrounds are opposite polarity (MIT Perceptual Science Group, n.d.).
3-D plausibility: add a casting object, consistent direction of light, and coherent occlusion cues for best results. Remove them and the illusion weakens (MIT Perceptual Science Group, n.d.).

Why it still matters

Checker–Shadow is more than a parlour trick; it’s a compact testbed for lightness constancy, showing that the brain’s best guess about materials and lighting can trump raw luminance. It connects low-level normalisation, mid-level segmentation/anchoring, and scene interpretation in one image—useful both for vision science and for graphics/vision modelling (Adelson, 2000; Gilchrist, 2006).

Bottom line

Square A and B are the same luminance; your brain relabels them because it segments the scene into surfaces and illumination and then discounts the shadow—with simultaneous-contrast nudges from their different surrounds (MIT Perceptual Science Group, n.d.).
The illusion thrives on soft shadow edges, plausible 3-D cues, and opposite local surrounds; break those and it collapses (MIT Perceptual Science Group, n.d.).
The success of the effect shows how good (and occasionally gullible) human lightness perception is when it tries to infer reflectance from context, not just read off pixel values (Adelson, 2000).