How Does Face Recognition Work?

The rest of the post (with images) is coming soon. 

Summary

  • Break the image down into sets of ever-smaller squares of pixels.
  • Analyze the average brightness of the set of pixels.
  • Compare the average brightness of a square to the squares around it and look for patterns called Haar features.
  • Combine “probability face = yes” into a final determination: Is this a face?
  • Some algorithms use color or motion (video) as well.
  • Anti-face makeup can help you avoid being detected.

Note:

Face detection is finding a face in an image. Face identification is matching the found face to a database and identifying an individual. People use the term “facial recognition” to cover either or both detection and identification.

 

The Basics

All algorithms (a concise way of saying a set of rules and patterns) follow a basic sequence: input, processing, output. Facial detection can seem almost mystical at times, so I like to start at the basics to help make things less intimidating.