From CV Dazzle, a project by Adam Harvey that demonstrates how makeup, hair styling, and accessories can be used to foil facial recognition technology. Harvey ran the below looks through a neural network that evaluates the similarities between two faces. Each face is given a score between 0 and 1, with scores below 0.8 interpreted as likely not a match. When compared with a makeup-free image of the same subject, the looks below (one to four) received scores of 1.0, 0.73, 0.56, and 0.53, showing the extent to which contouring and asymmetrical designs can be used to confuse the software by altering the spatial and textural relationships between key features such as the eyes, eyebrows, and nose. These are considered more identifiable than features in the bottom half of a face, which is dynamic and easily obscured by collars, masks, or other garments.

Previous

Next