The Facts on Facial Recognition with Artificial Intelligence
Amazon Rekognition makes it easy to add highly accurate image and video analysis to your applications. As with many technological advances, it’s important to understand how facial recognition works, and how it can be used. Let’s take a look at some of the most frequently asked questions.
What is facial recognition?
Facial recognition is a system built to identify a person from an image or video. This technology has been around for decades, but its usage has become more noticeable, and accessible, in the past few years as it now powers innovative solutions, such as personal photo applications and secondary authentication for mobile devices. To understand these emerging capabilities, let’s first discuss how facial recognition works.
Facial analysis capabilities, such as those available in Amazon Rekognition, allow users to understand where faces exist in an image or video, as well as what attributes those faces have. For example, Amazon Rekognition can analyze attributes such as eyes open or closed, mood, hair color, as well as the visual geometry of a face. These detected attributes become increasingly useful for customers that need to organize or search through millions of images in seconds using metadata tags (e.g., happy, glasses, age range) or to identify a person (i.e., facial recognition using either a source image or a unique identifier).
How are customers using facial recognition?
Facial recognition is useful across many applications and industry verticals. Today, we see this technology helping news organizations identify celebrities in their coverage of significant events, providing secondary authentication for mobile applications, automatically indexing image and video files for media and entertainment companies, all the way to allowing humanitarian groups to identify and rescue human trafficking victims.
Marinus Analytics, for example, uses artificial intelligence with Amazon Rekognition to provide agencies with tools, such as Traffic Jam, that assist them in identifying and locating victims of human trafficking. Investigators save invaluable time by using image analysis to search automatically through millions of records in seconds, which previously required individual analysis by investigators.
Another example is Aella Credit, a financial services company based in West Africa that provides banking services via a mobile app for underbanked individuals in emerging markets. Using Amazon Rekognition’s ability to detect and compare faces, Aella Credit can provide identity verification, without any human intervention. This simple use of facial recognition allows for more individuals to receive access to banking services than was ever previously possible. You can find other examples of customers using Amazon Rekognition here: Amazon Rekognition Customers.
How should I apply facial recognition responsibly?
Facial recognition should never be used in a way that violates an individual’s rights, including the right to privacy, or makes autonomous decisions for scenarios that require analysis by a human. For example, when a bank uses tools like Amazon Rekognition in a financial application to verify their customers’ identity, the bank should always clearly disclose their use of the technology and ask the customer approval of the terms and conditions. Regarding public safety and law enforcement, we think that governments are free to work with law enforcement agencies to develop acceptable use policies for facial recognition technologies that both protects the rights of citizens and enables law enforcement to protect the public’s safety.
In all public safety and law enforcement scenarios, technology like Amazon Rekognition should only be used to narrow the field of potential matches. The responses from Amazon Rekognition allow officials to quickly get a set of potential faces for further human analysis. Given the seriousness of public safety use cases, human judgment is necessary to augment facial recognition, and facial recognition software should not be used autonomously.
As stated by Dr. Matt Wood, “Machine learning is a very valuable tool to help law enforcement agencies, and while being concerned it’s applied correctly, we should not throw away the oven because the temperature could be set wrong and burn the pizza. It is a very reasonable idea, however, for the government to weigh in and specify what temperature (or confidence levels) it wants law enforcement agencies to meet to assist in their public safety work.”
How does facial recognition work in Amazon Rekognition?
Faces are matched based on their visual geometry, including the relationship between the eyes, nose, brow, mouth, and other facial features. When images are analyzed by Amazon Rekognition, there is an outline around the face, called a bounding box, which determines the only part of the image Rekognition considers in its analysis. The analysis then produces object notation numbers for the image that indicate the “location” for the major elements of the face. When customers are running a face search, the technology is comparing this data from the source image to each of the images it searches. From there, the service assigns each face in the image a similarity score. This approach ensures that Amazon Rekognition has no information about the identity of an individual, only the likelihood that one face is a potential match for another.
What are a similarity score and a similarity threshold?
A similarity score is a statistical measure of how likely two faces in an image are the same person, when analyzed by Amazon Rekognition. An image that received a similarity score of 95% for instance, would indicate that amongst all the faces Rekogniton analyzed, this image had a 95% similarity with the face being searched for. A higher similarity score means the more likely the two images are from the same identity. That said, even a 99% similarity does not guarantee it is a positive match.
That is because Rekognition uses what is called a probabilistic system, where determinations cannot be made with absolute precise accuracy, it is instead, a prediction.
This is where the similarity threshold comes into play. A similarity threshold is the lowest similarity score the application using Rekognition is willing to accept as a possible match. The choice of threshold has a fundamental impact on the search results that are returned. The number of misidentifications (sometimes called ‘false positives’) that can be afforded by the customer is a direct result of the threshold setting. A customer will select the appropriate setting based on their needs and use case of the application.
We recommend a 99% threshold setting for use cases where highly accurate face similarity matches are important. In public safety and law enforcement scenarios for example, this is often a key first step to help narrow the field and allow humans to expeditiously review and consider options using their judgment.
On the other hand, many scenarios don’t require human review of Amazon Rekognition responses. For example, secondary factor authentication with an employee badge and a face recognized by Amazon Rekognition with a high (99%) similarity. Or a personal photo collection application, where a few incorrect matches can be tolerated, a lower threshold of 80% may be acceptable. Customers can tune the similarity threshold to the specifics of their use case and needs.
What is the Celebrity Recognition API? Is that the same or different than doing a face search?
Celebrity Detection is designed to identify potential famous people across different movie scenes and environments. Since celebrities often play different characters (wearing different makeup, wigs, and other distortions to their appearance), this Amazon Rekognition feature has been trained on pre-labeled data to return the highest probable matches within a specific list of famous people. By design, this use case allows for a higher number of false positives and should not be used in public safety or law enforcement use cases.
In contrast, Rekognition’s Face Search feature is designed to tell you the precise amount of similarity between two faces – and it can be optimized for precise matches and used in security and public safety applications, such as finding missing children and reuniting them with their parents, authorizing employee access to a building, or identifying and rescuing victims of human trafficking.
These two features are completely different regarding the underlying technology they use, the use cases they solve, and the customers they serve.
Is facial recognition safe?
Yes. Let’s examine some common misconceptions about facial recognition and how it works.
First, some believe that people can match faces to photos better than machines. However, the National Institute for Standards and Technology (NIST) recently shared a study of facial recognition technologies that are at least two years behind the models used in Amazon Rekognition and concluded that even older technologies could outperform human facial recognition capabilities.
Second, as in all probabilistic systems, the mere existence of false positives doesn’t mean facial recognition is flawed. Rather, it emphasizes the need to follow best practices, such as setting a reasonable similarity threshold that correlates with the given use case. Also, one of the advantages of this technology is that it continuously learns and improves, so false positives can be reduced over time.
Today, many successful customers, such as Thorn, VidMob, Marinus Analytics, and POPSUGAR are using facial recognition in simple ways that have a powerful impact.
How can I get started with facial recognition?
AWS provides 10-minute tutorials and in-depth documentation with prescriptive guidance to help you get started using facial recognition.
How can I report potential Amazon Rekognition abuse?
If you suspect that Amazon Rekognition is being used in manner that is abusive or illegal, or infringes on your rights or the rights of other people, please report this use and AWS will investigate the issue.