Technical Details

How does facial recognition work?

Large-scale facial recognition is becoming increasingly prevalent in the digital age. Under large-scale facial recognition, we understand algorithms used to bulk-analyse picture data uploaded to social networks or stored in private surveillance systems.

Most modern facial recognition techniques use Neural Networks (NN) as their basis. This means that the algorithm is trained on a large dataset of faces of people where the corresponding name is known. The NN then learns to detect any face it has already seen from different angles, at different distances, different levels of image detail, or different levels of obstruction of the face (such as a blurry picture, someone standing somewhat in front of someone else, etc.).

In general, such an algorithm is heavily dependent on significant features in faces, such as the shape and size of eyes, mouth or nose of a subject, as well as clear boundaries of the face, such as the jawline.

To successfully detect a face, the database needs reference pictures of a person, as well as some name linked to them. Since “tagged” pictures like that exist for many of us on facebook, in newspapers, other social networks and similar services, most people with an internet presence will somehow be detectable by a facial recognition algorithm.

What are the dangers of large-scale surveillance systems?

Instances of abuse in the past and scenarios for abuse in the future.

How to solve this problem

Summary of a how to obstruct facial recognition in general.

Large-scale facial recognition can be obstructed in a couple of different ways. First off, you can make sure that no pictures of yourself appear on the internet in the first place. This becomes very hard or impossible if they are already online, or you have any job in which you need an internet presence. Second, you can blur or otherwise edit pictures of yourself before uploading them.

Our method

Description of how mask works.

Potential drawbacks

Emphasis on the fact that this does not shield from law-enforcement in any way, shape or form.