da/sec scientific talk on Biometrics

Topic: Towards Isolated Motion Blur Face Image Quality Assessment and Defect Classification

by Torsten Schlett
D19/2.03a (also online via the corresponding BBB room), December 12, 2025 (Friday), 12.00 noon

Keywords — Face Recognition, Quality Assessment, Motion Blur, Specific Image Defect Detection

Abstract

„Motion blur is caused by movement relative to the camera, which can affect face recognition performance. Existing ‚unified‘ (i.e. not defect-specific) face image quality assessment approaches can already estimate the face recognition utility of face images with varying motion blur defects, as shown e.g. in the previous da/sec scientific talk on this topic (2025-07-16). This may be sufficient to for instance discard lower quality face images during a capture process due to motion blur.

However, since these approaches by design do not exclusively correlate with motion blur defects in isolation, they naturally cannot be directly used to reliably differentiate between motion blur and other defects, including blur due to defocus instead of motion. Albeit perhaps comparatively less important than general utility assessment, the capability to assess different defect types in isolation could plausibly be useful to identify and possibly counteract concrete quality-affecting issues, e.g. by gathering statistics for different image capture sites.

This talk presents corresponding ongoing research, including an extension of a previously shown general utility assessment evaluation for motion blur data, as well as new defect-isolated parts.“