Facebook’s facial recognition technology has rapidly advanced over the years, raising questions around how the social media giant uses its ability to identify people’s faces in photos and videos. Here we’ll examine what facial recognition Facebook currently employs, what it’s used for, and how accurate it is.
How does Facebook’s facial recognition work?
Facebook builds a unique facial recognition template for each of its users by analyzing the photos and videos they’ve been tagged in. This produces a highly detailed map of the specific contours and features of a person’s face. When a new photo or video is uploaded, Facebook’s algorithm compares the faces it detects to its library of facial recognition templates. If there’s a match, Facebook may suggest tagging that person in the photo or video.
Facebook’s facial recognition system uses a DeepFace algorithm. This is a form of artificial intelligence trained through deep learning on a dataset of 4 million facial images. According to Facebook, DeepFace reaches an accuracy of 97.25% in detecting faces, compared to the 97.53% of the human brain.
How accurate is Facebook’s facial recognition?
Facebook claims DeepFace has near-human accuracy in facial verification tests. However, some studies suggest real-world results may be lower, influenced by factors like image quality, lighting, profile angle, and facial expression.
One key test is Facebook’s ability to match different photos of the same person, which one study found was correct around 83% of the time. Performance dropped for photos taken many years apart in age. Matching accuracy has likely improved somewhat since then with further training of the DeepFace algorithm.
Overall, for standard frontal photos in good lighting conditions, Facebook’s facial recognition is reasonably accurate when suggesting tags. But it can still make mistakes, especially with lower-quality images.
What does Facebook use facial recognition for?
Facebook uses facial recognition in a few key areas:
Photo tagging suggestions
When photos are uploaded, Facebook scans for recognizable faces and suggests tagging people who most closely match its facial templates. This makes tagging faster. Users can confirm or ignore these suggestions before posting.
Automatic alt text
Facebook’s AI automatically generates descriptions of photos for visually impaired users. Facial recognition identifies people to be mentioned in these alt text descriptions.
Identity verification
Facebook may use facial recognition to verify a user’s identity, for example when unlocking an account or making payments in Oculus VR headsets.
Facebook’s photo review system
Facebook claims facial recognition helps review reported photos that may violate its rules. It is not used to identify people for ad targeting.
Here is an overview of Facebook’s use of facial recognition technology:
| Use Case | Description |
|---|---|
| Photo tagging suggestions | Scans new photos and suggests people to tag based on its facial recognition templates |
| Automatic alt text | Identifies people in photos to describe to visually impaired users |
| Identity verification | Confirms user identity by matching faces to account photos |
| Photo review system | Helps review reported photos violating policies using facial recognition |
Can Facebook recognize you without being tagged?
Facebook may be able to recognize your face even if you’re untagged in photos or videos on its platform. That’s because it can create a facial recognition template of you from other sources, like:
- Photos where you’re tagged by name on Facebook
- Photos of you from a friend’s profile if they have tag review enabled
- Other photos or videos you may appear in organically on Facebook
If enough identifying images of you exist from these sources, Facebook can build up its own facial recognition template for you without your active participation. This template could then be used to identify you in new uploads.
How to opt out of Facebook’s face recognition
You can opt out of Facebook using your face for suggestions and features. To do this:
- Go to “Settings” then “Face Recognition”
- Toggle facial recognition to “Off”
This prevents Facebook from creating a facial recognition template with your identifying facial information. However, you may still appear in friends’ suggestions unless you delete tags they add of you.
Can Facebook detect faces it hasn’t seen before?
Yes, Facebook’s facial recognition technology is advanced enough to detect any recognizable face in an image, even people who don’t have a Facebook account. The algorithm looks for basic facial features like eyes, nose, and mouth to detect a face.
However, Facebook cannot identify who the person is or match them to a specific facial template if it hasn’t previously analyzed photos of that individual. Its facial recognition capabilities require building up a detailed map of a person’s facial characteristics to work.
How deeply can Facebook “see” your face?
Facebook’s facial recognition algorithm maps out the key contours of people’s faces in order to identify them. According to researchers, some of the facial features analyzed include:
- Distance between eyes
- Eye, nose and mouth shape
- Eyebrow thickness and shape
- Lines and wrinkles
- Jawline
However, Facebook’s system does not perform any deeper analysis of facial characteristics – it only focuses on what’s needed to match your photo to an existing template or suggest tags.
Facebook’s artificial intelligence cannot infer emotions, gender, ethnicity, personality or other attributes just by looking at your face in its normal use cases. Some researchers have managed to build algorithms that can guess certain traits from faces, but Facebook says it avoids doing this due to ethical concerns.
Should you be concerned about Facebook’s facial recognition?
Opinions vary on whether Facebook’s use of facial recognition is cause for concern. Here are some of the key considerations around privacy and ethics:
Privacy concerns
Facebook is amassing detailed facial scans on an unprecedented scale without always making users aware. This bothers privacy advocates, even if the data is currently only being used for tagging suggestions and not ad targeting.
Consent issues
Facebook builds facial recognition templates without explicit opt-in consent. Users may appear untagged in friends’ photos without realizing. Critics argue this lacks transparency.
Potential future misuse
While Facebook currently states facial recognition data won’t be used for ads or surveillance, many worry it could change policies in the future.
Inaccurate tagging
Facial recognition can sometimes misidentify people, leading to embarrassing or sensitive photos being wrongly tagged.
Police use concerns
Facebook has provided facial recognition templates to law enforcement in the past to assist investigations. But unclear rules around its use have raised accountability issues.
However, supporters argue that Facebook’s practices are reasonable for a technology that can enhance user convenience and safety.
Conclusion
Facebook’s rapidly improving facial recognition abilities allow it to identify users in photos and videos with a high degree of accuracy under good conditions. However, questions remain around privacy, consent, and the potential for misuse that set it apart from other biometric identification technologies.
Users uncomfortable with Facebook’s collection and use of facial templates can opt out of facial recognition features, albeit at the cost of losing functionality that can be convenient when used properly. Facial recognition technology will likely continue advancing and spreading to more uses, bringing ongoing scrutiny and debate over its place in social media.