Contextual contrast
Identical gray values can appear different when nearby luminance and texture change.
Truuts experiment
Perceptual CAPTCHA Experiment
A visual challenge where identical gray shapes appear different because equal-luminance surroundings are organized into different spatial contexts.
Quick facts
- Creator
- Eugene Truuts
- Project type
- Human perception and security research prototype
- Task
- Five contextual-contrast judgments
- Research goal
- Increase the economic cost of reliable automation
- Runtime
- Browser · React · HTML Canvas
- Status
- Experimental prototype · Updated August 2026
About the experiment
The experiment is a five-round visual game: in every challenge, choose the shape that appears darkest. All five organic forms use exactly the same base gray. Their contexts also contain equal areas of the same light and dark tones, so a global average or histogram produces the same answer for every option.
The illusion comes from arrangement. Light pixels concentrated near a contour change perceived lightness differently from the same pixels positioned farther away or split into alternating sectors. The final percentage describes how often your choices followed the intended contrast condition; it is not a binary human-or-bot verdict.
A useful CAPTCHA does not have to be literally unsolvable by machines. Its practical job is to make reliable automation economically unattractive: each solved action should cost an attacker more time, compute, model access or human labor than the action is worth. Perceptual variation can add friction, while one-time server challenges, rate limits, risk scoring and frequent challenge changes multiply that cost.
This remains a research prototype, not a claim of bot-proof verification. A production flow would require server-generated one-time challenges, continuous evaluation against current vision models and an accessible non-visual alternative.
Interactive demo
Complete five challenges by choosing the shape that appears darkest. The final report shows your match rate and explains what the result can — and cannot — tell us.
Click the identical-gray shape that appears darkest.
Technical concepts
Matched luminance
Every context uses equal-area light and dark annuli, defeating a simple comparison of mean brightness or histograms.
Spatial organization
Identical context pixels are rearranged radially or into balanced sectors, changing perception without changing aggregate values.
Reaction time
Response timing offers a behavioral signal, although it cannot establish identity on its own.
Five-round score
Multiple trials expose perceptual consistency and produce a more informative result than a single pass-or-fail click.
Economic deterrence
The objective is not perfect machine resistance, but raising the marginal cost of automated abuse beyond its expected value.
Post-answer inspection
The intended contrast condition is revealed only after a choice, keeping the mechanism explainable without leaking the answer first.
Security limits
Perceptual novelty is a hypothesis to test, not a substitute for rate limits, abuse controls and accessibility.