BSI is developing the Open Source Face Image Quality (OFIQ framework, which will enable transparent and detailed quality assessment of facial images in the future.

When dealing with operational biometric systems, the quality of biometric data plays an important role in ensuring the usability of the data in the reference databases and thus keeping the performance of the systems based on these reference databases high.

Both in Germany and internationally, fingerprints and face images are primarily used in public biometric systems. The applications of these biometric modalities continue to increase. For example, the “Entry-Exit System” (EES) is currently being introduced in the area of ​​European border control, in which face images and fingerprints of third-country nationals will be stored in the future. With a planned 200-300 million data records, the EES is an enormously large database, meaning that particularly high quality requirements are placed on the biometric data collected.

The use of biometrics in government areas is also increasing nationally. For example, with the “Live Enrolment Initiative”, which comes into force in May 2025, the scanning of passport photos brought with you in the German passport and ID authorities will largely be abolished. Instead, passport photos will (almost) exclusively be captured live in the future. The reliable quality assessment of the face images is becoming increasingly important in order to make their use as optimal as possible, for example when entering other countries.

Good and reliable quality assessment algorithms are therefore of crucial importance both nationally and at European or international level. At the same time, however, it is important that the quality assessment of biometrics delivers uniform and comprehensible results so that the quality of biometrics is understood equally in all components and processes of a system. As part of European border control around the EES, for example, it is essential that the biometric systems at all border crossing points in the Schengen area evaluate the quality of the recorded biometrics equally in order to prevent loopholes and inconsistencies. In order to achieve this goal, an open algorithm (open source) is necessary that provides transparent results on the quality of the biometrics under consideration.

While there has been a corresponding open source algorithm for fingerprints with the “NFIQ algorithm” (NIST Fingerprint Image Quality) since 2004, such an open algorithm for the area of ​​ face images is still completely missing.

The BSI is therefore currently developing the counterpart “OFIQ” ​​(Open Source Face Image Quality) to close this gap and enable an open and transparent quality assessment of face images both nationally and internationally.

Background

In addition to fingerprints, face images are one of the most widely used characteristics in biometric systems - both in commercial systems and in the public sector.

The quality of the face images in a biometric system directly determines the recognition performance of the entire system - the larger the database becomes, the greater the influence of poor quality on efficiency. The quality assessment of face images therefore plays a crucial role, particularly in very large databases, such as the upcoming Entry-Exit System (EES).

The quality of light images is determined by a variety of different factors, such as the uniformity of the lighting, the position of the head, the sharpness of the image and whether the eyes are open or closed.

In total there are more than 20 of these so-called “quality components”. Each individual component has a direct influence on the overall quality of the face images. However, it is currently hardly known how big the influence of the various components actually is. On the other hand, the effort required to design certain components well when capturing face images is sometimes high to very high. In order to be able to weigh up the effort and benefits against each other, the quality components must be further examined.

Based on the assessment of the individual quality components, concrete and usable feedback can be given to the recording and recording persons in the recording process: Instead of reporting that the quality of the face images is generally not sufficient, a precise statement can be made as to which factors, if any, can be improved and what influence this will have on the quality of the face images. This makes the capture process more accurate and delivers better results, which in turn improves the biometric performance of the overall system.

Standardisation

The international standardization on the subject of face images quality takes place via the ISO standard ISO/IEC 29794-5. The OFIQ framework is used here as a reference implementation and can be used in commercial and government applications in the future.

The OFIQ source code is available on GitHub.

Quality Component Testing

To support the development of ISO/IEC 29794-5, the National Institute of Standards and Technology (NIST) launched the FATE SIDD campaign to test light image quality in which the various proposed algorithms are tested per quality component.

You can also find further information about the testing procedure on the NIST website.

Further Information

The European Commission's Joint Research Centre (EU-JRC) identified the need for facial 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: 1) the combination of a number of individual values estimating human-defined features such as illumination, sharpness, pose, background, etc. 2) deep-learning derived features; or 3) a combination of both hand-crafted and deep-based features.“

Further information about OFIQ and the background information can also be found here:

• International Face Performance Conference (IFPC) 2022: Introduction OFIQ
• Draft ISO/IEC CD3 29794-5 (October 2023)
• State-Of-The-Art-Report  on the OFIQ (November 2022)
• Paper on the assessment of facial image quality: T. Schlett, C. Rathgeb, O. Henniger, J. Galbally, J. Fierrez, C. Busch: „Face Image Quality Assessment: A Literature Survey“, in ACM Computing Surveys (CSUR), (2022)