Clearview enters NIST biometric accuracy test with a splash amid image struggle

Clearview AI is taking steps to back up its biometric accuracy claims and show the legitimacy of its business model, with a strong entry into the U.S. National Institute of Standards and Technology’s facial verification ongoing benchmark.

The company’s CEO also explained to Biometric Update the choice of the specific NIST test the algorithm was submitted to.

The latest edition of NIST’s Face Recognition Vendor Test for 1:1 Verification shows results for Clearview among the top vendors for accuracy in the visa, mugshot, ‘visaborder’ and border categories, with matching accuracy consistently above 99 percent across all demographics. The company points out in an announcement that it achieved the highest match rate among American vendors in the ‘wild photos’ category, and second-highest in the world.

Other than the mugshot result, 11th in the field as of an October 28 update, Clearview’s algorithm achieved a top 5 result in each category.

The company says its facial recognition technology is used for rapid identification of suspects and criminal investigations, and that it has been used in thousands of cases to find missing children, identify people with dementia, and apprehend dangerous criminals.

Why Verification?

Clearview submitted its algorithm to NIST’s verification test, rather than the FRVT 1:N (Identification), despite the law enforcement app’s actual use being one-to-many matching.

“We submitted to the NIST 1:1 verification test for two main reasons,” Clearview AI CEO Hoan Ton-That explained to Biometric Update in an email. “First, when we started the submission process, NIST required a passing score on the 1:1 before submitting for 1:N. By the time that requirement was lifted in August, we already committed to the 1:1 submission. Second, the 1:1 test provides the only breakdown by racial demographic. We had the utmost confidence our algorithm would perform exceedingly well. With greater than 99 percent accuracy across all demographics, our confidence is validated. We will submit for 1:N verification soon. We look forward to sharing the results, when available.”

Ton-That also confirmed that the company’s platform is entirely cloud-based, and told Biometric Update that the company is “in the process of rolling out the algorithm tested by NIST.”

Perception push and pull

Clearview seems to believe it is on the right track towards turning public perception. Ton-That told an audience at an event held by the Federalist Society at Stanford University that with law enforcement successes using its facial recognition, “people are coming around” to the technology’s permanent place in society, according to a report by Mlex.

Examples Ton-That provided include the investigation of the January 6 attack on the U.S. Capitol building by the FBI and the 2019 identification of a child pornographer by the Department of Homeland Services.

The company has grown alongside these successes from around 10 employees early last year to 45.

Meanwhile a joint investigation by Information Commissioners in Australia and the UK have concluded, and while the UK’s ICO is considering its next steps, the Office of the Australian Information Commissioner (OAIC) has declared that Clearview breached the privacy of Australians.

The company’s actions violated Australian’s privacy by collecting their biometrics without consent and through unfair means, the regulator said, and through not notifying data subjects, ensuring their information was accurate, or taking reasonable steps to comply with the country’s Privacy Principles.

The OACI has ordered Clearview to stop collecting Australian’s facial images and biometric templates, and to destroy those it currently holds. The regulator rejected the company’s arguments that it was not handling protected personal information, and that as a U.S. company, it did not need to comply with the Privacy Act.

The OACI also suggested that online platforms should do more to protect the data they hold from being scraped by third parties.

Article Topics

accuracy  |  algorithms  |  biometric identification  |  biometric matching  |  biometric testing  |  biometrics  |  Clearview AI  |  Face Recognition Vendor Test (FRVT)  |  facial recognition  |  identity verification  |  law enforcement  |  NIST  |  privacy