Thanh Tung Linh Nguyen received the best poster award for the contribution of „NeutrEx-Light: Efficient Expression Neutrality Estimation for Facial Image Quality Assessment “ at the 23rd International Conference of the Biometrics Special Interest Group (BIOSIG 2024), which took place in Darmstadt, Germany.
Abstract:
Face recognition systems have numerous practical applications, including device authentication, border control, and attendance monitoring. These systems must achieve high recognition performance while processing large volumes of data. To ensure that low-quality images do not degrade biometric performance, it is crucial to quantify the quality of biometric samples. The draft international standard ISO/IEC 29794-5 introduces the concept of component quality, which measures the quality of individual quality elements. In this work, we focus on NeutrEx, a recently proposed component quality measure that quantifies the expression neutrality of facial images. We optimize NeutrEx to improve its efficiency in terms of parameters, storage space, and inference time. We investigate the applicability of optimization techniques, including pruning and knowledge distillation, to enhance the throughput rate of NeutrEx to tailor it for high-frequency real-world applications. All code and pre-trained experiments will be publicly available upon acceptance.