Foiling facial recognition

Cameras that recognise faces have become ambient, from the supermarket entrance to a neighbour’s video doorbell. None of these countermeasures make a face unrecognisable in every setting. They raise the cost of automated identification.

A face is a data source

Facial recognition runs in more places than most people register:

  • Retail analytics measure dwell time and repeat visits.

  • Public and transport cameras log routes, timestamped.

  • Social platforms auto-tag uploaded photos, often without asking.

Avoiding one platform does little on its own. Current systems identify people from side profiles, partial occlusion, and in some cases gait, so a single measure rarely closes the gap.

Infrared-blocking glasses

These resemble ordinary glasses but reflect infrared, which many camera systems rely on to locate and scan a face. Prescription versions exist, Reflectacles and IRpair among them. The effect is to wash out the eye region for the camera while staying unremarkable to a human observer. Overhead night-vision systems work on the same principle and can be disrupted the same way, with less predictable results.

Anti-surveillance clothing and masks

Masks and printed patterns can break the facial contours a detector looks for. Dazzle-style patterns disrupt contour detection; metallic threading interferes with thermal imaging. Combined with a hood and sunglasses, the aim is to deny the detector a clean front-facing capture rather than to disappear entirely.

Lower-cost alternatives

For a smaller budget:

  • Stick-on infrared LEDs flare the sensor with light the eye ignores.

  • UV-reactive face paint reads as distortion to some cameras while looking subtle in daylight.

  • A generic mask reduces capture quality, though it can draw attention of its own in some settings.

Limits

  • Gait recognition is untouched by anything worn on the face.

  • Some systems adapt, so a fixed method degrades over time; varying it helps.

  • Visible countermeasures can attract human attention even as they defeat the camera.

Fuller measures

For a higher threat model:

  • A thermal blanket masks body heat from infrared tracking.

  • Voice-altering tools address voice recognition, which is now a parallel channel.

  • Fabric with metallic lining blocks some overhead imaging.

Facial recognition depends on consistent, predictable input. Varying appearance across time and place is what degrades it.

Last reviewed: 2026-07-08.