New research on mobile fingerprint biometric scanning, synthetic data presented at eu-LISA summit
Among a series of presentations on technologies that are or could be used on Europe’s border at an eu-LISA Industry Roundtable held last week in France, two biometrics research projects, one involving hardware and another involving software, stood out.
Javier Galbally of eu-LISA chaired the session of industry presentations on ‘Biometric Solutions and Business Processes for Passenger Processing at Land/Sea BCPs,’ which began with a pair of presenters from the French Forensic Police and Isorg sharing their work on mobile biometrics capturing.
French police want mobile fingerprinting capacity in part to collect the biometric data of victims of crime, in order to differentiate it from other data, such as latent fingerprints found at a crime scene. Identification of cadavers, identity and right of residence verification and enrolment in information systems for minor offences also motivated the project.
Isorg Biometric Security and Identity Program Manager Jennifer Aflalo explained Isorg’s organic photodiode sensors, and the implementation of the sensor, and image quality considerations. Used in a scanner that provides a dedicated light source, Isorg’s technology meets the criteria for high-security fingerprint quality, Aflalo says.
Initial prototype design was followed by a larger-area version, and feedback collected from the French forensic authorities, and then a smartphone-sized system designed for final assessment and comparison to other market alternatives.
Good quality images were provided with minimal processing, marking a success for Isorg, according to the presentation, with similar results to other FBI-certified biometric technologies.
As an optical device, the scanner can also capture ID documents and even latent prints, Aflalo says. The tests showed that latent prints could be collected on the device, and then prints from live subjects, without the quality of images from the latter being affected for up to 100 repetitions.
One of the major advantages of the test from Isorg’s perspective was the work to put in place the image processing necessary to meet the requirements of the police.
Marcel Grimmer, a researcher with NTNU, presented a project conducted with eu-LISA to evaluate and improve the synthetic biometric datasets used by the regional agency.
The biometric synthetic datasets were composed to provide the necessary volume of training data without running afoul of data access and privacy challenges.
The test aimed to determine if the synthetic faces have similar characteristics to genuine samples, evaluate their quality and compare the scores obtained with open-source facial recognition systems. To do so, the researchers evaluated synthetic non-mated face images, edited facial attributes to generate synthetic mated samples, and compared the mated faces to real data.
Generative adversarial networks (GANs) were used for modifications like changing apparent facial age.
Grimmer reviewed the rapid progress in the capabilities of these GANs, setting up the finding that the synthetic subsets each have similar biometric quality to the real reference data.
The mated comparison exhibited a much lower degree of similarity than the non-mated comparisons, particularly with relative extremes in age.
Grimmer concludes that face age modification algorithms appear to be effective for generating training data, in terms of biometric quality. Further experiments on robustness to larger age differences are recommended.
Some challenges must be solved before the synthetic data is fit for purpose as a sole tool for assessing operational systems, despite the encouraging results, Grimmer says.