Pixel-Level Face Image Quality Assessment for Explainable Face Recognition

In this work, we introduce the concept of pixel-level face image quality that determines utility single pixels in a for recognition. We propose training-free approach to assess qualities given an arbitrary recognition network. To achieve this, model-specific value input is estimated and used build sample-specific regression model. Based on model, quality-based gradients are back-propagated converted into estimates. experiments, qualitatively quantitatively investigated meaningfulness our proposed based real artificial disturbances by comparing explanation maps faces incompliant with ICAO standards. all scenarios, results demonstrate solution produces meaningful enhancing interpretability its quality. The code publicly available.