How accurate is facial recognition?

Accuracy depends on the algorithm, image quality, and type of comparison being performed. Verification against one high-quality reference image can be extremely accurate under controlled conditions, while searching a large database using poor-quality surveillance footage is more difficult. NIST testing has found large differences between facial recognition algorithms. More accurate systems generally produce fewer errors and smaller demographic differences.

Is Face ID facial recognition?

Yes. Phone face unlock systems are a form of facial recognition, but they generally perform one-to-one verification. The system checks whether the person using the phone matches the face enrolled by its owner rather than searching a large database to identify an unknown person.

What is a face finder?

A face finder is a tool that uses facial recognition to search for other images of the same person. You upload a photo, and the service compares the face with images in its index to look for possible matches. Face finders such as PimEyes may help locate where someone appears online, but results can also expose personal information or make people easier to identify without their knowledge.

Can facial recognition identify someone from one photo?

Potentially. A facial recognition system can use a single photo as the starting point for a search, but the result depends on the image quality, the algorithm, and whether the person appears in the database being searched. A possible match shouldn’t automatically be treated as proof of identity. Facial recognition systems can return false matches, particularly when images are unclear or searches involve very large databases.

Can you opt out of facial recognition?

Sometimes. Some consumer face-search services offer opt-out tools, and apps or devices may let you disable facial recognition features. Opting out of one service doesn’t remove the original photographs from the internet or stop other companies from collecting them. Reducing public access to your images and removing them from source websites provides broader protection.

Artificial Intelligence (AI) > Facial recognition

Facial recognition explained: When your face becomes a commodity

Facial recognition is used to unlock phones, verify identities, control access, and identify people from photos or video. The same technology can also search huge image databases or recognize someone in a public space. Here’s how facial recognition works, where it’s used, and what it means for your privacy.

What is facial recognition?

Facial recognition is biometric technology used to verify or identify people from images or video. Depending on the system, it may confirm someone’s identity, identify an unknown person, or find other images of the same person.

Those uses have very different privacy implications:

Table
Type	What it does	Privacy implications
Face verification	Checks whether a face matches one stored identity	Usually involves a face knowingly enrolled for a specific purpose
Face identification	Compares a face against many stored identities to find a possible match	Identifies people who never actively agreed to be searched
Face search	Uses a photo to find other images of the same person across an image index	Connects an unknown face with websites, profiles, and other personal information

How does facial recognition work?

Facial recognition measures features of a face and converts them into biometric data points for comparison. For example, a smartphone compares those points with the ones recorded when you set up face unlock, while a face search system may compare them with data from millions or billions of other images. The exact process varies, but usually follows a few basic steps:

Capture the face: A camera, photo, or video provides an image containing a face.
Detect and prepare it: Software locates the face and adjusts the image so it can be analyzed.
Create a biometric template: Characteristics of the face are converted into a mathematical representation, sometimes called a faceprint or facial embedding.
Compare the face: The template is compared with a stored template or a database of other faces.
Calculate a match: The system produces a similarity or confidence score. If the score is high enough, the system treats it as the same person.

Facial recognition before and after AI

Computerized facial recognition dates back to the 1960s. Early systems relied on researchers identifying key facial features, such as the eyes, nose, and mouth, then measuring the distances and relationships between them so a computer could compare one face with another.

AI and machine learning changed that process dramatically. Modern systems analyze faces automatically, detect far more complex patterns, and compare images at enormous scale. Facial recognition can now search huge databases and potentially connect a face captured in public with images and information collected elsewhere.

Where facial recognition is commonly used

Facial recognition is now used across everyday services, commercial spaces, and public institutions for everything from convenience and security to investigation and surveillance.

Table
Common use	How	Why
Phones and accounts	Compares your face with a previously enrolled template	Unlock devices and verify account access
Banking and fraud prevention	Confirms that someone matches an identity or account holder	Prevent impersonation and fraudulent transactions
Airports and borders	Compares travelers with passport photos or other identity records	Automate identity and security checks
Law enforcement	Searches photos or video against databases of known faces	Generate investigative leads or identify people
Stores and venues	Recognizes visitors for security, access, or personalization	Prevent theft, control entry, or tailor services
Advertising and personalization	Links a recognized face with a customer profile	Personalize offers, displays, or marketing
Face search engines	Searches an online image index for other images of the same person	Find where someone appears on the web

The privacy risks of facial recognition

Facial recognition could expose far more than your identity. It makes it easier to track, profile, identify, or connect people with information collected elsewhere.

Face search identifies strangers

Face search turns a single photograph into a way to identify someone. There are many online services designed to search online images for matching faces, potentially revealing names, workplaces, social profiles, and other personal details. Clearview AI provides similar technology to law enforcement and government customers.

Combined with data brokers or people-search sites, those matches could become a starting point for stalking or doxxing(new window). Researchers have also shown how facial recognition and smart glasses could identify strangers and retrieve personal information about them in real time.

Facial recognition tracks where you go

Combining cameras with facial recognition makes it possible to follow people across public spaces without requiring them to identify themselves. At protests or transport hubs, facial recognition may reveal who was present. Across larger camera networks, it could track movements, reducing anonymity and potentially deterring people from attending sensitive events.

Private organizations use the same technology to control access. Madison Square Garden Entertainment, for example, used facial recognition to identify and deny entry (new window)to lawyers working for firms involved in litigation against the company.

Your face is a commercial identifier

Facial recognition can link someone entering a store or venue with an existing customer profile. Businesses may then use that identity for security, personalization, loyalty programs, or targeted marketing.

Trakomatic, a Microsoft partner, developed an opt-in retail system(new window) using Microsoft Azure that recognized shoppers as they entered a mall and used their profiles to personalize offers, digital displays, and services.

Wrong matches have serious consequences

Facial recognition systems don’t perform equally well in every situation. Accuracy varies with the algorithm, image quality, camera angle, and type of search, while the US National Institute of Standards and Technology (NIST) has also found demographic differences in false-positive rates.

Facial recognition errors have already led to wrongful arrests and false accusations. Robert Williams was wrongfully arrested in Detroit after facial recognition incorrectly matched him to a suspect, while the FTC said Rite Aid’s system(new window) generated thousands of false-positive matches.

Old photos could end up in facial recognition systems

Photos uploaded to social media, news sites, or other websites may be scraped to train AI and used for facial recognition, even though identification was never the reason they were originally published.

A photo shared years earlier may end up in a facial recognition system the person in it never expected to join. In 2021, the FTC alleged that Everalbum(new window) used millions of facial images extracted from users’ photos to help develop its facial-recognition technology, despite misleading people about when facial recognition would be applied.

Biometric data is difficult to replace

Passwords can be changed after a breach. It’s harder to change your face. Stolen biometric data may continue to pose a risk long after the original security incident.

In 2019, large parts of the database behind the Biostar 2 biometric access were left unencrypted(new window). It exposed more than one million fingerprint records, along with facial recognition data and images of users. People affected by a breach like that can’t simply replace their face or fingerprints in the way they would change a compromised password.

Pushback is coming, slowly

Governments are starting to place limits on facial recognition. Under the EU AI Act, AI systems are not permitted to create or expand facial recognition databases by indiscriminately scraping faces from the internet or CCTV footage. Real-time facial identification by law enforcement in public spaces is also generally banned, with limited exceptions for situations such as finding missing people, preventing serious threats, or identifying suspects in certain crimes.

The US has no single nationwide law governing facial recognition. Instead, states have taken different approaches. Washington and Colorado require safeguards around some government uses, while Maine bans most government use and allows police to request searches only under strict conditions. At least 18 states have considered legislation regulating police use of facial recognition in recent years.

How to reduce your exposure to facial recognition

You can’t prevent every camera from capturing your face, but you can make a facial match less revealing by reducing the personal information and photos available about you online.

Limit who can see photos of you on social media and review old public accounts.
Remove unwanted photos from websites at the original source whenever possible.
Use opt-out tools provided by services such as PimEyes to stop your face appearing in their search results.
Remove personal information from data brokers and people-search sites like TruthFinder so a facial match reveals less about you.
Review facial recognition settings before enabling biometric features in apps or online services.
Avoid uploading facial images to services unless you understand how they will be stored and used.

Switch to a private AI assistant

Lumo is an AI assistant designed for people who want the benefits of AI without giving another company more personal data to collect. Your conversations are not used to train AI models or retained in processing logs, while saved chats are protected with zero-access encryption.

Try Lumo now

Frequently asked questions about facial recognition

How accurate is facial recognition?: Accuracy depends on the algorithm, image quality, and type of comparison being performed. Verification against one high-quality reference image can be extremely accurate under controlled conditions, while searching a large database using poor-quality surveillance footage is more difficult.
NIST testing has found large differences between facial recognition algorithms. More accurate systems generally produce fewer errors and smaller demographic differences.
Is Face ID facial recognition?: Yes. Phone face unlock systems are a form of facial recognition, but they generally perform one-to-one verification. The system checks whether the person using the phone matches the face enrolled by its owner rather than searching a large database to identify an unknown person.
What is a face finder?: A face finder is a tool that uses facial recognition to search for other images of the same person. You upload a photo, and the service compares the face with images in its index to look for possible matches.
Face finders such as PimEyes may help locate where someone appears online, but results can also expose personal information or make people easier to identify without their knowledge.
Can facial recognition identify someone from one photo?: Potentially. A facial recognition system can use a single photo as the starting point for a search, but the result depends on the image quality, the algorithm, and whether the person appears in the database being searched.
A possible match shouldn’t automatically be treated as proof of identity. Facial recognition systems can return false matches, particularly when images are unclear or searches involve very large databases.
Can you opt out of facial recognition?: Sometimes. Some consumer face-search services offer opt-out tools, and apps or devices may let you disable facial recognition features.
Opting out of one service doesn’t remove the original photographs from the internet or stop other companies from collecting them. Reducing public access to your images and removing them from source websites provides broader protection.