Blog Articles

Understanding the Technology Behind Facebook's Face Recognition

  • By Rana
  • 27-02-2018
  • Software

Some users came across Facebook’s newly introduced face recognition feature. Questions like how Facebook could detect a face and match it with a similar other from among over 2 billion users have been roaming around. Facebook claims its new technology is just as accurate as the human-eye!!

What Does this Feature Mean to You as a User?

This face recognition feature will notify you if a picture of you was uploaded by a friend of yours or a friend of a friend. It will also detect if someone uses a picture of yours, whether you know them or not, as their profile picture. Facebook intends to make people feel more confident with the data they upload to Facebook. A strategy Facebook is following to show its users that they care about their privacy, an often seen poll by Facebook in our homepages. A counterargument is that Facebook does not exactly care about your privacy, but it wants your data for itself.

Human Intelligence VS Artificial Intelligence

As human beings, we can recognize a famous person from the back. If you saw only Obama’s forehead and hair, you would instantly know it’s Obama. Facebook aims to develop its AI technology the same way.

If a human was asked if two pictures of unfamiliar people belong to the same person, this human can get it right 97.53%, where Facebook’s newly developed software can get 97.25% right of the same challenge, regardless of variations in the picture or whether the person was looking or not.

What Technology is Facebook Using?

The DeepFace technology which is introduced by Facebook is claimed to be just as accurate as the human eye; it consists of a nine layer neurons which sees the creation of connections between those neurons. Each image uploaded to facebook is studied carefully by this technology, then every image gets a unique fingerprint at the bottom of the layers of the neurons.

This new artificial intelligence software uses this network of simulated neurons to learn to recognize patterns in huge amounts of data. The software learns to recognize patterns in digital images just like a human being.

This software analyzes the neurons or pixels in each photo a person is tagged in, then generates a string of numbers that is known as a “template”. When a photo is uploaded, it is compared with other templates and then stored within its rightful template. This is how Facebook manages to collect a database for every person registered.

What Challenges did Facebook Face?

A challenging problem was the variations in photos, including lighting, clothing, glasses, different poses, different camera angles, image resolution, etc. Only 52% of people have frontal face pictures that are fit for facial recognition. With this large percentage of pictures unfit for facial recognition, Facebook had to enhance its technology and find a solution.

The Solution

Facebook solved this problem by creating the Pose Invariant Person Recognition (PIPER) method which relies on subtle cues from the body, such as hair style, clothes, glasses, pose and other contexts. These are recognized as “poselets”. Poselets are classifiers that detect common pose patterns and pose variations. Face recognition is one of the many things PIPER stores. This makes it easy to recognize people even if their face is not visible.

 

Artificial Intelligence is taking on every big firm. Mark Zuckerberg, Facebook’s CEO, stated "One reason I'm so optimistic about AI is that improvements in basic research improve systems across so many different fields — from diagnosing diseases to keep us healthy, to improving self-driving cars to keep us safe, and from showing you better content in News Feed to delivering you more relevant search results. AI is going to make our lives better in the future.”

However, this still begs the question, with Facebook’s huge database and intelligence its creating, can Facebook’s Artificial Intelligence become dangerous one day?