How to Evaluate a Biometric Facial Image

FIQA or Face Image Quality Assessment is the process of evaluating the quality of a facial image by considering factors such as sharpness, lighting, pose, and facial expressions to ensure efficient performance in facial recognition systems. (Athar et al.) This evaluation is important for improving the performance of systems such as facial recognition, biometric authentication and medical imaging. The consequences of poor quality images extend beyond performance drops. For example, in medical imaging, inaccurate interpretation can impact diagnostic decisions. Ensuring that images meet a perceptual quality threshold is therefore essential for accuracy and security.

While automated systems increasingly rely on facial imagery, there remains a gap between objective quality assessment and human perception. This gap can lead to system failures, highlighting the urgent need for more human-aligned quality metrics.

This project investigates whether fusion-based objective metrics can better approximate subjective facial image quality compared to traditional standalone measures. For that, we use a dataset of facial images with controlled distortions and collect subjective quality scores from human evaluation. We then apply various objective image quality assessment metrics and test them, and the fusion of them through a set of approaches, to see which of them better approximates to subjective perception. This has the goal to bridge the gap between objective evaluation and human perception and judgement.