Pareidolia: Faces in the Static

Noise Field

Noise scale / roughness 4.5

Contrast / detection threshold 0.50

Detector Output

Candidate blobs

Faces detected

Top confidence

Seed

–

Detected face-like arrangement (score ≥ threshold)
Dark blob = candidate "eye"

How it works

Pareidolia is the tendency to perceive a specific, meaningful pattern — most famously a face — in random or ambiguous visual stimuli: clouds, wood grain, scorched toast, a shadow on a wall at night. It arises because human vision has dedicated, highly tuned face-detection circuitry, centered on the fusiform face area (FFA) in the ventral temporal cortex, which responds preferentially and very rapidly to any arrangement of two roughly symmetric dark regions above a third feature — the generic "two eyes, one mouth" template. That template is so broad and so eagerly triggered that it readily fires on coincidental noise that only loosely matches it, producing a false-positive face percept with no real face present.

This over-sensitivity is thought to be adaptive rather than a flaw: across evolutionary history, the cost of a false positive (seeing a face that isn't there) is trivial, while the cost of a false negative (missing a real face — a predator, a rival, a hidden observer) can be lethal. Natural selection tunes a detector's threshold to the asymmetry of its error costs, so a system with such lopsided stakes settles on a low, trigger-happy threshold — exactly the bias this simulation lets you dial in directly.

The overlay in this simulation runs a simplified, genuine version of that template match: it scans the noise field for local minima (dark blobs), pairs up two that sit at plausible "eye" spacing and height, checks the region below the pair for a contrasting "mouth" feature, and scores the whole arrangement — exactly the coarse blob-and-geometry logic believed to underlie the brain's first-pass face detection, before slower, higher-resolution cortex confirms or rejects it.

scale = 4.5 · threshold = 0.50