Home Office identifies areas for improvement in biometrics self-enrollment systems
Thumbs and eyeglasses are notable challenges for biometric self-enrollment solutions, the UK Home Office has found, and informed the audience at the latest European Association for Biometrics (EAB) event.
Home Office Head of Technical Architecture for Biometrics Graham Camm and Senior Policy Advisor for Future Biometrics Jordan Webster presented findings from the first phase of trials of biometric self-enrollment for immigration processes in the EAB Lunch Talk.
The scheme for non-visa visitors, planned for launch next year, is similar to other electronic travel authorization systems around the world, Webster explains. It is expected that the scheme will eventually be used to process between 27 and 30 million people each year, and move beyond face biometrics to include fingerprints.
Remote biometric self-enrollment is the focus of the project, primarily with smartphones but also with unsupervised kiosks.
Those requiring visas are considered higher-risk, and therefore will likely have their biometrics enrolled in a supervised environment.
Feasibility trial findings
The 2021 feasibility trials for face and fingerprint biometric matching and presentation attack detection with volunteers from the public were divided into six categories of technologies, Camm explained. Idemia was the only supplier to submit technologies for all six. Half a dozen technology providers participated in the trial of fingerprint PAD on a personal device, despite it being “more novel” than some other tasks Camm said.
Binding face biometrics to an ID document seemed to work fairly well, but including fingerprints in that binding process proved more challenging. Camm says only one provider “really understood the problem space correctly.” Even that provider acknowledged that the technology to bind fingerprints to an ID document is in its infancy.
The kiosks submitted by suppliers were mostly similar in their design, according to Camm, but there was more variance between the various mobile solutions proposed. Many of the mobile solutions providers come from the identity verification space, Camm notes, and the results indicate the significant difference between that application and biometric enrollment.
The trial results showed that all solutions providers across both capture device types met the efficiency target of three minutes or less for enrollment. User satisfaction was likewise high for all kiosks. For mobile solutions capturing face biometrics, the user satisfaction standard of 95 percent was met by the best-in-class, but lower generally. For mobile fingerprint capture, user satisfaction was lowest, with particular difficulty noted for capturing thumbprints.
The effectiveness of the capture technology was also generally considered unsatisfactory, with too many enrollment attempts triggering review processes, such as for image quality. A high number of reviews would add cost, and ultimately make the system unsustainable.
A combination of face PAD solutions was able to identify all attack species for both kiosks and mobile devices, and like capture on kiosks, Camm speculates that improvements in tuning could yield better results. Fingerprint PAD effectiveness was considered insufficient.
No supplier was able to provide assurance that the face and fingerprints being bound to an ID document were from the same person, Camm notes.
The kiosks met the biometric matching accuracy target, but only the best-in-class mobile solution met the target for face, and none did for fingerprints.
Eyeglasses were the most common cause of low scores, and apps did not remind people to remove their hats or masks, which negatively impacted some attempts.
Matching scores, determined using the latest available Idemia face matching algorithm, were slightly higher for kiosks than for mobile solutions, with the best of the former scoring a 99.1 percent TPIR at 0 percent FPIR, while the latter class was led by a score of 98.5 percent TPIR at the same threshold.
Some of the contactless fingerprint capture technologies delivered prints that look indistinguishable from prints captured with a contact scanner, Camm says. Others, however, distorted or even introduced (false) minutia.
Camm described some of the challenges that remain in contactless fingerprint capture based on Home Office observations for the EAB audience.
Home Office also considered latent-mark search performance, demographic differentials, and usability.
The next phase of testing was conducted separately for kiosks and mobile devices, due to the differences in maturity between the two.
The kiosk test began in early 2022, and fingerprint app testing will launch in late 2022.
Home Office considers the greatest risk for the second round of feasibility trials to be the chance that they provide no new insights, which would then require another round of trials in late-2023 or early-2024. To mitigate the expense those would impose on both Home Office and participating suppliers, smaller ‘benchmark tests’ are planned to gage the maturity of suppliers’ solutions.
Once the benchmarking exercises are successful, a larger trial which includes subjects as young as five years old will be conducted.
Future tests may consider palm biometrics capture, enrollment with a smartphone assisted by an acquaintance of the subject, and further focus on usability.