Customs and Border Protection (CBP) continues to get match rates between 97 and 98 percent with its use of facial comparison technology where it is deployed to track the entry and exit of foreign nationals and U.S. citizens arriving to and departing from the U.S. and the agency isn’t finding that the technology is discriminating based on individual demographics, an agency official said on Thursday.
“We’re not seeing demographic-based” issues in the 2 to 3 percent of instances where a photo of someone taken at an entry or departure point doesn’t match with the facial image of the individual in a government database, John Wagner, deputy assistant executive commissioner for CBP’s Office of Field Operations, told the House Homeland Security Committee.
Wagner also said he’s not aware of any instances of false positives, which would mean someone that went through the facial comparison check at entry station or departure gate was let through because the technology wrongly identified a match against a photo in the database.
The committee’s hearing on the Department of Homeland Security’s use of facial recognition and other biometric technologies followed the release in December of a government report that found many face recognition algorithms do a poor job in being able to accurately match against images contained in a database due to age, gender, race and country of origin. However, the National Institute of Standards and Technology’s (NIST) evaluation found that in the case of the highest-performing algorithms used in finding a match of a photo against a large number of images contained in a database, the demographic differentials were negligible (Defense Daily, Dec. 19, 2019).
Asked by Rep. Bennie Thompson (D-Miss.), the committee chairman, about algorithms misidentifying black men and Asians, Charles Romine, director of Information Technology Laboratory at NIST, answered that with “the highest performing algorithms we don’t see that to a statistical level of significance for one-to-many algorithms.”
Wagner said that given CBP is using an algorithm from one of the highest performing vendors evaluated by NIST, “we’re confident that our results are corroborated with the findings of this report.”
After years of trial and error and seeing advancements across the board in biometric technology, CBP in the last year of the Obama administration began to move forward with a deliberative approach to use face recognition technology at international airport departure gates to verify the departure of foreign nationals in accordance with their visa terms. The technology is meant to replace manual checks with increased accuracy and efficiency.
Congress, at the behest of the 9/11 Commission, directed the use of biometric technology to track the entry and departure of foreign nationals into and out of the U.S.
CBP has been using fingerprint technology for around 15 years to track entries but only in the past few years has found a way to track departures. However, given limited real estate and operational confinements, CBP found that it works better to also use the face recognition technology with U.S. citizens, who may opt out of using the technology and use legacy identity checks.
CBP is also rolling out face comparison technology to the entry process and applying it to foreign nationals and U.S. citizens.
The face recognition algorithm that CBP is using in their Traveler Verification Service (TVS), which hosts the backend photo galleries for people arriving to and leaving the U.S., is provided by Japan’s NEC Corp. NIST’s evaluation found NEC’s latest algorithm to be one of the high-performing technologies.
CBP is currently using an earlier version of the NEC algorithm and plans to switch to the latest version of the algorithm, called NEC 3, in March, Wagner said.
In addition to algorithm quality, other factors are important in obtaining more accurate match rates and obtaining higher percentages of overall matching.
Wagner acknowledged during the hearing that operational factors such as lighting, human behavior, camera quality, the size of the database, the number of images of a particular individual in the database, and the quality and age of the image or images in the database can impact matching results. He said cameras must meet CBP standards and the agency is working with stakeholders to ensure other factors are optimized to further improve matching.
Wagner said that CBP has worked with its stakeholders on all of these variables to ensure the conditions for high match rates. Airlines and airports are buying the camera systems being used at airport departure gates and integrating the biometric checks with their existing processes so that a face match with CBP’s TVS means individuals don’t have to bother with presenting their passports and boarding passes as is typically required.
Asked by Thompson how to get to get to zero the number of people CBP fails to match, Wagner said given that the agency has the person’s image in the backend gallery, “we should have matched them.”
Wagner pointed out that the image galleries contained in the TVS are relatively small, several thousands of images, versus millions, which helps CBP improve its matching.
CBP has signed a memorandum of understanding with NIST for NIST to evaluate the operational variables the agency is using with its biometric checks to measure the impact of these factors, Wagner said.
Prior to the hearing, an electronic privacy group, the Electronic Privacy Information Center (EPIC), set a letter to Thompson and Rep. Mike Rogers (R-Ala.), the ranking member on the committee, calling for the panel to ban DHS from using face recognition technology.
EPIC wants the U.S. and other countries to first “establish the legal rules, technical standards, and ethical guidelines necessary to safeguard fundamental rights and comply with legal obligations before further deployment of this technology occurs.”
Thompson has generally supported the use of biometric technology in government applications but has concerns related to privacy and data security.
“I want to reiterate that I am not wholly opposed to the use of facial recognition technology, as I recognize that it can be valuable to homeland security and serve as a facilitation tool for the department’s varying missions,” he said at the outset of the hearing. “But I remain deeply concerned about privacy, transparency, data security, and the accuracy of this technology and want to ensure these concerns are addressed before the department deploys it further.”
Wagner pointed out that CBP’s use of face recognition at land, air and seaports is not for government surveillance purposes, pointing out that the cameras aren’t hidden and the technology is used to verify a person’s identity against their passport.