Face biometrics image quality assessment tool maturing as eu-LISA plans integration

The Open Source Face Image Quality software library is intended to support large-scale biometrics programs with information about the usefulness of photos for biometric comparison. OFIQ is a work in progress as those systems roll out, therefore, despite being  the reference implementation for ISO/IEC 29794-5 Biometric sample quality – Part 5: Face image data. The maturity of OFIQ and the ongoing work to improve it was examined in an online presentation hosted by the European Association of Biometrics (EAB) on Tuesday, in cooperation with the German Federal Office for Information Security (BSI) and eu-LISA.

It was the second OFIQ user group meeting, following one held in mid-2025. At that meeting, the user community discussed the biometric data quality assessment tool’s role in border control and other government applications, the OFIQ Demonstrator and the standard’s future, including a possible version 2 in 2027.

OFIQ is the best performing algorithm for predicting low facial recognition confidence scores based on comparisons of NIST SIDD (Specific Image Defect Detection) Error versus Discard Characteristic (EDC) curves, Christoph Busch of Hochschule Darmstadt explained during the introduction.

A study published weeks ago by the U.S. Department of Homeland Services based on research conducted by the Maryland Test Facility suggested OFIQ’s utility is “extremely limited” for DHS use cases.

The specific criteria used in the DHS report, however, including the size of the dataset and the selection of face biometrics systems from 15 vendors, make it less applicable to the goals of the user group, Busch argues.

Incremental improvements

Eu-LISA took a role in maintaining OFIQ last year. OFIQ is not yet integrated with the biometric systems eu-LISA manages. But the agency is currently planning to make the tool available for use with its systems in the foreseeable future, eu-LISA Capability Building Officer for Research and Development Javier Galbally told meeting participants. On the way to that point, eu-LISA recently introduced an FAQ for OFIQ and related resources to its website.

Version 1.0.2 of OFIQ, released last April, introduced fixes for several errors found in previous versions, Benjamin Tams of secunet explained. He does not recommend using earlier versions, therefore. Version 1.0.3, released in July, introduces the interface method “vectorQualityWithPreprocessingResults.”

Version 1.1.0 came out in October, and adds support for mobile platforms, various runtime improvements and the use of libjpeg-turbo to ensure consistent interpretation of pixel data when compiling code with different methods. Version 1.1.1 is coming soon, he says, with a library dependency update and other relatively minor changes.

Tams invited contributions to the OFIQ project, though not in the form of new quality measures, changes of output changes or configuration.

Secunet’s Maxim Schaubert discussed Python bindings and web services for OFIQ.

Coming soon: OFIQ 2.0

The company’s Johannes Merkle delved into the goals for OFIQ 2.0, including improved accuracy and robustness, computational efficiency and additional quality checks, such as for motion blur.

The SSD face detection algorithm is being replaced by CenterFace, which Merkle says is both faster and more effective. Secunet submitted CenterFace to NIST’s FATE Quality evaluation, in which “it performed extremely well,” according to Merkle.

The PPFL landmark estimation algorithm may be replaced by ADNet to improve performance speed, though a final decision has not yet been made. Face parsing algorithm zllrunning could also be replaced with wizz4rt to improve processing speed.

Once preprocessing is complete, OFIQ 2 will have a new background uniformity algorithm, which Merkle says provides several advantages, and has performed well in both NIST FATE Quality and internal evaluations. The new version’s illumination uniformity algorithm, under-exposure prevention, which can be a factor in performance for people with darker skin, and head pose evaluation are also being upgraded.

A promising candidate for improved motion blur analysis has been submitted by a student, trained on sharp images, some with motion blurring, and others with non-motion blurring. Algorithms are also being developed to assess when the subject is looking into the camera.

The third OFIQ user group meeting is scheduled for July 20, 2026.

Article Topics

biometric data quality  |  biometrics research  |  EAB  |  EAB 2026  |  eu-LISA  |  face biometrics  |  ISO 29794-5  |  open source  |  Open Source Face Image Quality (OFIQ)