As the first blog in this series explained, the term “face recognition technologies” refers to both face verification (1:1 comparison) and face identification (1:N comparison) technologies. These technologies are distinct from face detection and face analysis technologies.
How do different types of face recognition technologies work? And how do users in the public and private sectors deploy them?
This second blog in the series on face recognition technology governance challenges will address those questions and explain why distinguishing between functional applications and use cases can support efforts to advance face recognition technology legislation.
Challenge 2: Many Functional Applications
In many still image face recognition technology applications, a camera takes a photograph that face recognition technologies use to generate a “probe” face template. The face recognition technologies then compare the probe template to one (in the case of face verification) or several (in the case of face identification) face template(s) in a gallery.
Outputs vary depending on the face recognition system’s design. If the probe template is similar enough to a gallery template to clear a preset threshold, some face recognition systems simply indicate that the probe template represents someone in the gallery. For example, the system could convey that the probe template represents an authorized individual by unlocking a device, displaying a check mark, or opening a gate. Other face recognition systems return the information stored with the gallery template (such as the individual’s name and/or photograph) that is most similar to the probe template and that clears a preset threshold. These systems convey which specific individual in the gallery the probe template represents, not just that the probe template represents someone in the gallery.
The face identification systems that many in law enforcement use to help generate investigative leads compare the probe template from a still image to a gallery of face templates and return a preset number of results (often called a “ranked candidate list”). These ranked candidate lists consist of the information stored with the gallery templates that were most similar to the probe template and a percentage score measuring the similarity between the probe template and each gallery template. For example, a law enforcement agency may configure a face recognition system to produce a ranked candidate list of ten entries. This system would use a probe template from a photograph to query a face template gallery and return a list of the ten names and photographs (along with similarity scores) associated with the gallery templates that were most similar to the probe template.
Real-time video face recognition technology applications work similarly, except that, instead of generating a probe face template from a still image, they generate one or more probe templates from faces detected in video footage. Often, these applications use face identification technologies to compare each probe template to a gallery of authorized (in the case of “trusted group” systems) or unauthorized (in the case of “watchlist” systems) users. Doing so can help determine whether the people in the video footage are authorized to be in a particular location.
In watchlist applications, if one or more probe templates is sufficiently similar to a watchlist gallery template to clear a preset threshold, the system may issue an alert simply indicating that someone from the watchlist may be in a secure location. Alternatively, the system may return information about the specific person on the watchlist whose gallery template was similar to the probe template.
In trusted group applications, the face recognition technology may compare a probe template to templates in a gallery of “trusted users.” Doing so can help track how many trusted individuals and non-trusted individuals entered a location, generate a log of which trusted individuals entered a location, and/or issue an alert if someone in a given location is not a member of the trusted group.
Additionally, technology vendors can incorporate face recognition technologies into systems that also perform other forms of biometric recognition, elevated body temperature sensing, scene analytics, payments, and other functions.
Challenge #3: Many Use Cases
As BPC previously explained, different users deploy various face recognition technologies and functional applications in a multitude of ways. State and local law enforcement and other government agencies use still image face identification technologies to help generate leads in criminal investigations; help identify missing, unconscious, or deceased individuals; and help combat fraud and identity theft. State government agencies also use face verification technologies to help control access to devices and locations.
U.S. federal law enforcement and other government agencies use still image face identification technologies to help check identity documents at border security checkpoints, generate investigative leads, and identify and locate missing and exploited children and other crime victims. Federal government agencies also use real-time video face identification technologies to scan “live camera feeds . . . for individuals on watchlists or suspected of criminal activity, which reduces the need for security guards to memorize these individuals’ faces.” Still image face verification technologies help federal officers and other federal government employees facilitate secure access control to physical and online locations.
Private sector use cases also abound. Individuals and organizations use still image face recognition technologies to unlock personal devices, automate photo tagging and file sorting, access private records, enter venues without having to show tickets, make cashless payments, provide contactless access control to office buildings and other secure facilities, check digital identification, and aid blind and low-vision individuals. Museums use real-time video face recognition technologies to provide visitors with customized entertainment experiences, and other private-sector entities may use real-time video face recognition technologies to bolster building security.
New use cases are likely to emerge as the face recognition technology market continues to grow in the United States and around the world.
Learning about face recognition technologies’ numerous functional applications and use cases necessitates a significant time investment, but the returns can be high. Differentiating between different functional applications and use cases can help policymakers tailor legislative requirements to the specific risks and benefits that different face recognition technologies can produce in different settings.
Understanding how accurate face recognition technologies are and how well they perform under various conditions, which the next piece in this series explains, can also help policymakers develop legislation that mitigates risks without unduly limiting benefits.
Support Research Like This
With your support, BPC can continue to fund important research like this by combining the best ideas from both parties to promote health, security, and opportunity for all Americans.Give Now