UCCS facial recognition dataset creator says research succeeded and privacy concerns misplaced
A biometrics researcher at the University of Colorado at Colorado Springs is setting the record straight about a facial recognition dataset created to help the military identify terrorists, The Colorado Springs Gazette reports.
The UnConstrained College Students (UCCS) dataset was a product of a $3.3 million government research project, and UCCS El Pomar Chair of Innovation and Security Professor Terrance Boult says it has been the subject of widespread confusion. The dataset is not available to the public, as many apparently thought.
The data is only available under a restricted-research license, but some people worried about its privacy implications are unclear on that part, Boult told The Gazette. “Part of the response is, ‘But it’s on the internet.’ My response is, ‘But it’s not.’”
The dataset, created in 2012 and 2013, has faced criticism in the past, but no biographic data is included with the images, and Bolt waited for five years from when the images were collected to make the UCCS dataset available. The professor notes that some people expressing concern about their privacy are doing so on Facebook next to pictures they have posted of themselves.
“The data was never given to the government, nor do we think they ever accessed it,” Boult said. “They paid us to do the research, and much like cancer research, they did not get all of the material.”
Initial funding for the project came from the Office of Naval Research’s Multidisciplinary University Research Initiatives Program, and a half-dozen other universities around the country have been involved.
The dataset consists of 1,700 images matched to at least one more of the same person with high enough quality for a confident match, from about 16,000 images of 2,400 different people. The images were collected in public, and their collection was reviewed by the UCCS Institutional Review Board.
Boult acknowledged a need to avoid situations in which a mistake results in a person being interrogated by the police, and said that was not possible from the UCCS project.
According to Lata Nott, the executive director of the First Amendment Center at think-tank Freedom Forum Institute, photography in public places as an expressive act is protected, algorithm training may not qualify.
The research done by UCCS and the other institutions reduced errors on controlled data by two to three orders of magnitude, according to Boult.