Facial recognition algorithms are designed to help people, like law enforcement agencies, identify otherwise difficult to determine characteristics in facial images. Sometimes this is meant to associate them with other similar faces, like a suspect, among other uses. Pareidoloop, a coding project from Phil McCarthy, certainly falls under one of those “other uses.” The project places randomly generated polygons over the top of each other and runs the result through facial recognition code until it produces an eerily recognizable face.
The variables can all be set by the user. It outputs an image of the created face at whatever size they want. It also measures up to a certain fitness for the image, which seems to be how likely it is that it’s a face, with a max of 35. The lower the fitness number, the faster it pops up a suitable image. If made too low, the entire thing just looks like a jumble of shapes. The number of generations the image goes through before it decides to start looking for another face, should it not hit the fitness required, is also able to be set. Essentially, users determine how much like a face it looks, how long it takes, and how big the image.
The whole process supposedly works better, and faster, in Chrome but the project itself is minimalist and doesn’t say much on the matter. It exists solely to provide haunting imagery of your creation. Set your variables, hit commence, and just let it go.
- This Japanese surveillance system can scan 36 million faces in a second
- Wasps are pretty good at recognizing faces as well
- Facebook uses facial recognition software on photos