Avoiding Facial Recognition

Always wear a face mask!

Types of Camera

One core thing to know about the types of cameras that are used to perform FR is that there are two distinct types or modes of FR. Many cameras are capable of both modes, but can only be in one mode at once. These modes are typical CCTV "visual-light", and IR-reflecting "night-vision". The reason a distinction must be made between these is that but

Visual-light cameras

Visual-light cameras can only see what the eye can see (±15nm) so they are easier to fool because you can "debug" your appearance and see what the camera sees with very little effort. This makes it easy to sufficiently cover-up yourself.

Night-vision cameras

Night-vision cameras can, despite their name, be engaged in full-daylight scenarios, however in practice this is rarely done because it slightly degrades the quality of the returned video and wastes energy illuminating an already bright area with non-visible light. Night-vision cameras can see through certain materials that visual-light can't, such as certain plastics, silicates, glass windows, and importantly eyeglasses.

Thermal cameras

Thermal cameras as surveillance for FR are not yet deployed widely because they are far too inaccurate, however the research into this area is coming along quickly. In 5-10 years this may become a serious threat, as thermal cameras can see through far more than night-vision cameras, including cloth masks.

FR and you

Hats and FR

Low-rim/high-cover hats (such as sunhats) are potentially one of the best tools to avoid FR because the heavy majority of surveillance cameras are placed well above the average eye-level, often mounted into ceilings ("eye-in-the-sky" camera types that are domes that poke down from building and train ceilings) or high points on walls ("arm-mounted surveillance cameras") for higher coverage area. The tradeoff here is that if the cameras can't see your face through the hat, you can't see anything above your eyeline either, as you must keep the hat pulled down to your own eyeline. This is also not a silver bullet, but best combined with some of the other techniques lined out in this page because there are plenty of cameras at eye level too.

Masks and FR

Face masks have been shown to be very effective at fooling the best FR systems. According to a US government survey on millions of photos of immigrants, even the very best of their 90-or-such FR algorithms tested had 5-50% error rates when the image presented had a mask on.

Not all masks are created equal. It seems that light-colored masks (particularly white ones) are far easier for algorithms to cancel out, while dark-colored masks (particularly black and very-dark brown) are extremely effective at confusing FR algorithms. It is potentially also noteworthy that similar-skin-colored face masks can potentially confuse visual-light FR far more than any other type, however this color category was not included in the US government survey.

Clothes and FR

Some modern FR systems often times take into account more context than just the head. Many top-shelf solutions in deployment by some police departments around the world will also use your outermost clothing layer to match you, so wearing the same outermost layer every day is recommended against, as this makes it easier for both algorithms and officers presented with possible matches to confirm positive FR matches. For example, wearing a single coat all the time that has a distinctive appearance will make it far easier for certain algorithms to identify you.

Hair and FR

It turns out that long hair is actually excellent for thwarting FR, not only because it blocks viewing your face from some angles but because of how much hair moves about and varies all the time. Long hair is a plus for avoiding FR.

Eyes and FR

The eye measurements, similar to the ones taken to develop eyeglass prescriptions, as well as measurements around the eye socket are some of the main measurements used to identify faces. However, this is difficult to act against, because IR and non-IR cameras will scan the eye socket in different ways, and a single pair of glasses can only block at most one of the two (except for Reflectacles, which offer some very expensive glasses that claim to block both, but I don't dare spend that much money to confirm). It should also be noted that many FR systems can compensate for the partial darkening effect that sunglasses "shades" provide, so these are not a reliable silver bullet fix. Because the nose and bridge are also key points, using traditional sunglasses is not a sufficiently effective method for avoiding FR alone, but darkened glasses/IR-blocking glasses can be very helpful as one tool of a set of tools used to stay anonymous.

Tilt and FR

Lateral tilts of 20+ degrees away from any given camera will dramatically reduce its effectiveness. Vertical tilts have not been reviewed in the same study but are likely to be less effective comparatively.

Motion and FR

Motion blur and lack of focus has also been shown to help reduce the risk of FR making matches against you, but since this would require sprinting everywhere all the time, this is potentially not super practical.

Extreme centralisation of CCTV networks

In the same way that every Ring doorbell is sending everything it sees to Amazon, pretty much every newly-installed CCTV camera nowadays is connected to the internet and uploading all the footage it records to one of the handful of American or Chinese companies that manufactured the camera. It is also worth noting that most of these cameras also have microphones built-in.

Datasets

Have you ever noticed that most ticket machines in Britain contain a camera that is perfectly positioned to capture nothing but your face? Make sure to smile as you tap your bank card!

Facial recognition works best when there's a clean dataset of photos specifically of just your face tied to your government ID, Driver's License, etc (or identifiers that are only a degree of separation away, like bank cards or phone numbers). Passport photos are an obvious source to check against, but consider how the following elements in everyday society will be helping to build this collection of photos of your face:

• Cameras in cash machines/ATMs (extremely common!)
• Cameras in train ticket machines (also extremely common!)
• Having your face captured when you pass through train stations
  • Cameras are positioned intentionally to get a clear image of everyone going through the gates to "prevent fraud" or whatever.
  • When using paper tickets, you can always ask staff to let you keep your ticket on the way out. They'll open the gates for you.
• Video calls through platforms such as Google Meets and Microsoft Teams
  • https://en.wikipedia.org/wiki/Optic_Nerve_(GCHQ)