Representing a face as a set of measurements

- [Voiceover] So far, our face recognition pipeline…can find faces in an image, extract them,…identify face landmarks, and then align the faces.…Now we're ready to dive into the most important step…in our face recognition pipeline:…telling faces apart from each other.…Here on the left, we have a face we want to identify…which we'll call the unknown face.…We also have a set of faces we already know…called the known faces.…We want to figure out which known face…is the best match for our unknown face.…The simplest approach would be to compare the unknown face…with each of our known face images.…

Whichever image has the most pixels in common…must be the best match.…However, this simplistic approach has two big problems.…First, if we have lots of known faces…directly comparing every pair of images would be too slow.…Second, just comparing the images themselves…isn't really the same as comparing the structure…of each person's face.…As a human, you can tell two people apart…even if they look very similar…because you recognize differences…