Biometrics in Public Sector Applications
Hinweis: Die internationalen Standards/Spezifikationen für Reisepässe werden von der ICAO (UN-Ebene) bzw. auf EU-Ebene festgelegt und sind in englischer Sprache formuliert. Beiträge des BSI, die in diesem Bereich Berücksichtigung finden sollen, müssen daher ebenfalls auf Englisch formuliert sein (vgl. hierzu auch die entsprechenden englischsprachigen Technischen Richtlinien auf der BSI-Website TR-03110, TR-03121, TR-03129, TR-03135, etc.). Deswegen finden Sie den folgenden Ausschreibungstext nur auf Englisch vor.
Biometric systems are utilized in a variety of domains ranging from user verification (e.g. access control or automated border crossing) to criminal investigations. Thereby, a range of biometric modalities such as facial images, fingerprints or iris images are utilized. As a fundamental principle and independently of the used modality, every biometric system compares a probe of a biometric modality with prior enroled reference images or templates.
Biometric systems utilize a comparison algorithm to pass a judgment on whether probe and reference (which was acquired previously) match. The better the quality of both, the probe and the reference, the more reliable the comparison score allows to distinguish a false match and a true match i.e. the false match and false rejection rate are minimized.
Thus, in domains where the quality of biometric samples can be influenced to some extent, system engineering is driven by enhancing the quality of captured samples. Typically, capabilities to positively influence the biometric capture processes exist e.g. by user guidance in scenarios such as application for official identity documents at a town hall or self-service biometric systems at border control.
However, if trying to optimize the quality of captured samples, e.g. by repeating a bad quality capture attempt, the necessity emerges to easily and reliable assess the quality of any sample in terms of applicability for comparison. If the applicability of a captured sample for reliable comparison can not be measured easily and accurately, improving the quality is difficult in large scale applications.
Aim and Scope
Within the scope of a thesis, algorithmic approaches to determine the quality of facial images should be considered. As of today, no single valued quality metric for biometric facial images exists. The goal of a thesis could be to suggest, develop and test metrics which are capable of assessing a captured facial image as suitable or unsuitable for biometric comparison. As a starting point, ISO  standardized different approaches to determine a facial image's applicability for comparison. However, no single valued metric is provided. Second, NIST  standardized a single valued metric to score a fingerprint capture's applicability for comparison.
The latest state of literature regarding approaches to assess biometric quality and corresponding metrics shall be reviewed. Thereby, approaches to measure quality of other modalities than facial images, could be considered in regard to their transferability. Unsupervised and supervised machine learning techniques could be reviewed for their adaptability.
The following exemplary issues in the domain of facial image quality metric could be tailored and addressed in a thesis, depending on the scope:
- How could available metrics for biometric facial image quality be combined to a single valued quality score?
- Could variations in the design of facial image comparison algorithms require dedicated training or tailoring of a facial image quality algorithm?
- How could machine learning techniques help to develop a single valued quality metric for biometric facial images?
 ISO/IEC 19794-5:2005 “Information technology - Biometric data interchange formats – Part 5: Face image data”
 Towards NFIQ II Lite Self-Organizing Maps for Fingerprint Image Quality Assessment NIST Interagency Report 7973
Type of Thesis
Master- or Bachelor-thesis or Diploma (tailoring of scope), English language is preferred but German language is possible.
Field of Study
Students with computer science, business informatics, mathematics, physics or statistics background with an affinity for digital security or image processing.
Markus Münzel, Florian Rienhardt, Markus Nuppeney,
Bundesamt für Sicherheit in der Informationstechnik,
Referat D 13: Kontrollinfrastrukturen und -architekturen