Interactive demo based on Simonotto et al., Phys. Rev. Lett. 78, 1186 (1997). Each animation frame adds independent zero-mean Gaussian noise to every pixel, then threshold-filters to black or white.
The paper models vision as a threshold crossing: each pixel's gray value I receives independent Gaussian noise j with standard deviation s. The displayed value is white if I + j > D, otherwise black. A new noise sample is drawn for every pixel on every frame, producing time-varying noise that the visual system can integrate — revealing structure invisible in a single static frame.
Stochastic resonance (sensory neurobiology) on Wikipedia — background on the phenomenon. Default image: Arc de Triomphe (source).