da/sec scientific talk on Biometrics

Topic: Motion Blur Face Image Quality Assessment

by Torsten Schlett
D19/2.03a, July 16, 2025 (Wednesday), 12.00 noon

Keywords — Face Recognition, Quality Assessment, Motion Blur

Abstract

„Motion blur is a type of blur caused by movement relative to the camera, and strong motion blur in face images can naturally affect face recognition performance.
This talk presents ongoing research about face image quality assessment for this specific defect type, based primarily on an internal face image dataset with various instances of real motion blur.
More specifically, this work concerns the quality assessment of single face images, as opposed to the joint assessment of multiple video frames.
‚Error versus Discard Characteristic‘ evaluation results indicate that existing ‚unified‘ (i.e. not defect-specific) face image quality assessment approaches already yield good results, so that further research specific to motion blur could potentially instead focus on 1. more computationally efficient approaches specialized to motion blur, and 2. the automatic distinction between motion blur defects and other image defect types, to provide actionable feedback.“