BSI is developing the Open Source Face Image Quality (OFIQ) Framework, which is intended to enable transparent and detailed quality assessment of facial images in the future. For the upcoming revision of the ISO standard ISO/IEC 29794-5, which provides and specifies algorithms for the quality assessment of facial images, the OFIQ framework represents the software-side implementation (reference implementation of the algorithms). The OFIQ framework should be applicable for various biometric applications, for example in border control scenarios.

In phase one many algorithms were initially adopted from the draft revisions of ISO/IEC 29794-5. Conversely, the algorithms themselves also contributed to revisions of the standard to be published. The various quality metrics and the recommendations and requirements of other relevant standards were also taken into account. The algorithms from the first phase were then improved based on the results obtained. The best algorithms were selected and defined as quality components.

Suitable test sets were collected and created for the evaluation. The (real) data contained therein was labelled for quality control purposes (test sets with ground truth labels). The algorithms were then evaluated in terms of their accuracy in predicting the data and its labelling (ground truth prediction). Algorithms that target quality components that could potentially affect the accuracy of face recognition are evaluated using EDC (Error-versus-Discard Characteristic) curves. This method evaluates the quality or efficiency of the quality assessment algorithms by quantifying how efficiently the discarding of samples with low quality values leads to improved error rates. Several open source and commercial facial recognition algorithms are used in this evaluation. The EDC evaluation aims to demonstrate the usefulness of the algorithm in the quality assessment of facial recognition. Where possible, the most promising algorithms were also subjected to the SIDD (Specific Image Defect Detection) track of the Face Analysis Technology Evaluation (FATE) Quality. This is an ongoing evaluation of software conducted by the US National Institute of Standards and Technology (NIST) that checks for quality problems in face images. Quality assessment algorithms are evaluated on their ability to assign low quality scores to border crossing images that are variously of non-ideal pose, illumination and resolution. NIST evaluates these algorithms using their own sequenced test data. Only algorithms relating to quality components covered by FATE Quality and the SIDD track were submitted.

OFIQ 1.0 - Final Project Report

This report documents the selection, implementation, evaluation and improvement of the individual algorithms that were implemented, tested and selected for OFIQ. Here you find the current versions of the Final Project Report:

• Report
• Short Report Open Source Face Image Quality (OFIQ) – Short Report
• Conformance Test Report Open Source Face Image Quality (OFIQ) – Conformance Tests