Street Wear to Stealth Wear : Low-tech Solutions to High-tech Surveillance

Inspired by the increasingly ubiquitous forms of surveillance and the recent Hong Kong protest, my project attempts to answer the question:

What easily accessible resources can be used to evade facial-recognition software?

Existing works have been done on objects ranging from clothing to jewelry. I’d like to highlight the works of two individuals: Adam Harvey’s hyperface pattern and Ewa Nowak’s Incognitio mask.

I wanted my solution to also account for the following limitations:

• While solutions such as facial projection (link) and camouflage makeup (link) can fool computers, they could easily be detected by passersby

• As a result of the Hong Kong protest, “masks” ( Guy Fawkes mask being the best known) have been banned from use

• Not all situations require the use of anti-recognition solutions. A versatile solution should be quickly removable or appears to be part of the “normal”

With that in mind, below are the iterations of my project.

Iteration 1: “garage door face mask”

Using the structure of a face protector, this first iteration builds on the Guy Fawkes mask idea, where a more realistic mask is velcroed to the revolving visor. When not in use, the mask can rest on the top of the head, ideally under some type of hat to be totally discreet.

Obviously, this iteration is more recognizable on the street, depending on how realistic the mask is. Since hyper-realistic masks are not easily accessible, this iteration builds the foundation for future improvements.

Iteration 2: “webs and mesh”

Using the outer shield of the face protector, different types of mesh materials are explored and tested for their ability to mask identity. In the end. every single panel displayed on the shield is able to deflect facial mapping / recognition.

The 4 panels on the left are from craft metal mesh, each painted to a different pattern or color. The finding is that patterns are not needed to conceal facial contours. The 2 center panels (pink, tan) also do the job adequately. So a user can paint the mesh to their own skin color to be more discreet.

The 2 large sections on the right are from honeycomb patterned paper and web styrofoam (the kind that wraps fruits). What’s interesting about the paper is that transparency is directional - the content can be totally visible when viewed from one angle but blocked from another angle. Since the natural folds have random distribution of angles, this material successfully blocked facial recognition. To the far right is the painted styrofoam web - a softer mesh that can be expanded - it also was able to block facial recognition.

Similar to the previous iteration, the mesh screen would be able to stored up when not in use.

This iteration improved upon the previous one by redefining the idea of “mask” to include “screen” that can be better manipulated and made less obvious to the naked eye while not sacrificing functionality against recognition software.

Iteration 3: “hat accessories”

The idea for this iteration came from the last two panels of the second iteration. What if the mesh is stretchable so that when it’s expanded, it shields user’s identity, but when retracted, it’s not noticeable? Extrapolating this further - softer and thinner - we arrive at a hat accessory that can be draped across the face when needed and be tied up as decoration otherwise.

This iteration is more suitable for women’s hats, as they tend to have wider rims and additional decorations that would help blend in. If the mesh cloth used is more similar to skin color, perhaps it will be less noticeable while in use. Still, this evokes of the funeral wear image and thus closer to the “normal” than the other iterations.

Iteration 4: “full mesh screen”

Chosen as a “winner” from the second iteration, this is a skin-color full mesh screen attached to a hat. When worn, it is harder to tell from a distance that someone is wearing a mesh. But it still tricks facial recognition cameras. If the mesh is adapted into a softer material - such as cloth - this could incorporate the advantages in the third iteration as well.