Project Sponsors

Bundesamt für Sicherheit in der Informationstechnik (BSI)

Project Partners

secunet Security Networks AG (SEC)

Contact

Christoph Busch
Email: christoph.buschh-da.de

Face Image Quality Assessment

Face Image Quality and Impact

When dealing with operational face recognition systems, the quality of biometric samples plays an important role to ensure the usability of the collected data in reference databases. Accuracy is highly dependent on the sample quality achieved during the capture process. For face image samples the quality can be expressed as a unified quality score and in various quality components such as illumination uniformity, focus, eyes open, frontal pose and many more. Measuring these quality components can provide actionable feedback in the capture process to the data subject or to the biometric attendant. Improving these quality components can lead to improved quality which in turn leads to better biometric performance.

Face Image Quality Standard

The outcome of the Open Source Face Image Quality (OFIQ) project is an open-source framework and reference implementation of face image quality assessment algorithms as standardised in ISO/IEC 29794-5, which can be deployed in commercial and government applications. The source code of OFIQ will be provided in GitHub.

Following the international standard ISO/IEC 29794-1, the quality of biometric samples is described with unified quality scores. The quality score is a holistic measure for the entire sample. The quality score is predictive of recognition performance and is an integer number in the range 0 to 100 (with higher being better). In addition, quality components (e.g. the pose angle or expression neutrality) are assessing details of the biometric sample. Components represent measurements on the biometric sample that may contribute to the computation of a unified quality score and/or facilitate actionable feedback.

Quality Component Testing

In order to support the development ISO/IEC 29794-5 the National Institute of Standards and Technology (NIST) has established the FATE SIDD face image quality test campaign, which is testing the various proposed algorithms for a quality component. SIDD is an abbreviation for Specific Image Defect Detection testing. The test plan is available here. On the NIST website you can find the current NIST FATE SIDD report (i.e. NIST.IR.8485).

Operational Deployment

The European Commission Joint Research Centre (EU-JRC) has identified the need for face image quality assessment in its 2019 study as follows: "... we recommend to promote the development of a vendor-independent, robust and reliable, face quality metric to be integrated in the ABIS-Face as soon as it becomes available. This quality metric could be the result of:

• the combination of a number of individual values estimating human-defined features such as illumination, sharpness, pose, background, etc.
• deep-learning derived features
• a combination of both hand-crafted and deep-based features
""

Further Information and Publications

• The OFIQ project website at BSI
• The OFIQ software in GitHub
• The OFIQ Python Adapter in GitHub
• The Draft International Standard ISO/IEC DIS 29794-5 (March 2024)
• The International Standard ISO/IEC 29794-1 biometric sample quality framework (April 2024)
• The NIST FATE SIDD face image quality testing
• Quality measure paper Facial Expression Neutrality (September 2023)
• Considerations paper on the Evaluation of Biometric Quality Assessment Algorithms (March 2023)
• The State-of-the-Art report from the OFIQ project (November 2022)
• Introduction paperFace Image Quality Assessment: A Literature Survey (September 2022)
• The iMARS project on Morphing Attack Detection and Face Image Quality (September 2020)
• The EU-JRC study on face image quality (July 2019)

Recorded Talks on Face Image Quality

• Presentation slides and recordings Face Image Quality Workshop November 2023
• Presentation slides and recordings IFPC Face Image Quality Session November 2022
• Presentation slides and recordings Face Image Quality Workshop November 2021