OFIQ report outlines evaluation techniques for algorithm accuracy and reliability

The Open Source Face Image Quality (OFIQ) project, led by the German Federal Office for Information Security, has published a preliminary report detailing its evaluation methodology. The report outlines crucial steps and approaches to ensuring the algorithms’ accuracy and dependability for measuring biometric image quality.

The software library is a reference implementation of the ISO/IEC 29794-5 international standard. The development and assessment of the OFIQ algorithms were conducted in two phases – initial implementation and assessment, followed by refinement and enhancement.

During the initial phase, the initial algorithms were implemented for a preliminary evaluation to assess their performance in facial image quality. The second phase involved refining and improving the algorithms based on the results of the first phase.

The evaluation process started with the collection and development of test sets with ground-truth labels. Algorithms are assessed based on their accuracy in predicting ground-truth labels. For numerical labels, empirical cumulative distribution functions (ECDFs) are used, while binary labels use detection error tradeoff (DET) curves.

Error-versus-discard characteristic (EDC) curves are plotted to assess how quality factors affect facial recognition accuracy. The purpose of the EDC assessment is to demonstrate the effectiveness of the algorithms in evaluating facial recognition quality.

The best-performing algorithms were subjected to the specific image defect detection (SIDD) track of the face analysis technology evaluation (FATE) Quality by NIST. According to the report, some commercial quality assessment algorithms were tested using the ground truth test sets.

Earlier this year, Biometix released a biometric face analysis engine, part of BQAT, the Biometric Quality Assessment Toolbox, which is based on the OFIQ algorithm.

Article Topics

biometric data quality  |  biometric testing  |  biometrics  |  face biometrics  |  Germany  |  ISO 29794-5  |  ISO standards  |  Open Source Face Image Quality (OFIQ)