Our website uses cookies so we can analyse our site usage and give you the best experience. Click "Accept" if you’re happy with this, or click "More" for information about cookies on our site, how to opt out, and how to disable cookies altogether.

More Accept ×

We respect your Do Not Track preference.

Close ×

opc header logo

Resources and learning

Print | Email this page

Rule 7 examples

A woman with short, curly hair flips through a notebook while browsing in a retail store. These scenarios are examples of how an agency might apply rule 7 in context.

Facial recognition by a retail store to operate a watchlist

A store uses FRT to identify individuals on a watchlist of previously violent customers to help improve staff and customer safety. An individual was flagged as being on the watchlist and asked to leave the store. The individual requests that the store corrects their biometric information by deleting their information from the watchlist (correction requests can include asking that information be deleted). How the store corrects the information could change depending on whether and why correction is needed:

  • If the individual should not be on the watchlist at all (e.g. because the store accidentally enrolled the wrong person onto the watchlist), the store would need to completely remove the individual’s biometric information from the watchlist.

  • If the individual was not on the watchlist, but was misidentified due to a false positive (incorrect match), the store would need to assess whether to:

    • Add a note in the system about past false positives/incorrect matches that would appear if the system identifies the person who is enrolled on the watchlist, in order for a staff member doing a manual review to have more context. This could also be appropriate for situations like if an individual has a twin or sibling who looks very similar.

    • Regenerate a biometric template for the person who is meant to be on the watchlist using a better-quality sample (e.g. CCTV still), if the enrolled photo is poor quality and contributed to the misidentification. Doing this would also help comply with the accuracy requirements in rule 8.

Receiving a correction request because of misidentification is a good prompt for the store to consider other changes to ensure the overall accuracy of the system and meet their obligations under rule 8. Some of the changes that the store should consider are:

  • Reviewing the system’s match threshold settings if the sensitivity setting is set too low and so the system is producing incorrect matches at too high a rate.
  • Removing the relevant biometric information for the correct person on the watchlist if the enrolled photo is not of sufficient quality to avoid further misidentifications for the person requesting correction.
  • Changing the process for responding to alerts to reduce the risk of subsequent misidentifications, for example by strengthening the manual staff review process (see also our rule 1, rule 5 and rule 8 guidance).

A black man, shown from the shoulders up, stands at a slight angle to the camera with a blank expression. A red laser line runs from his forehead to his chin, as though his face is being scanned. Facial recognition to control access to restricted site

A company is using FRT to control access to a restricted site. A staff member who should have access to the restricted site is repeatedly rejected by the FRT system. The staff member believes their biometric information needs to be corrected to resolve the issue. The company investigates and determines that the original biometric sample (photo of staff member) was not high enough quality to give consistent accurate results.

The company responds to the request by completely deleting the staff member’s existing biometric information. They then generate a new biometric template from a new biometric sample (photo of the individual).

 

 

 

 

Back to top