Peepers & Pixels: Human Recognition Accuracy on Low Resolution Faces

TL;DR

This paper quantifies how human face recognition accuracy degrades as probe-image IPD shrinks, identifying a practical reliability threshold around 10px IPD where accuracy nears chance. It combines a large, controlled dataset with standardized IPD and brightness adjustments to generate 320 impressions across five IPD levels, and uses untrained participants to emulate forensic review conditions. The results show a strong decline in human accuracy with decreasing IPD, yet humans maintain high decision certainty and react faster at low resolutions. In contrast, a state-of-the-art AFR (ArcFace) generally outperforms humans at higher IPD, but at the very lowest IPD a human examiner can outperform the algorithm, highlighting opportunities for human–algorithm collaboration in surveillance-era investigations. The findings inform guidelines for operational use of AFR, emphasizing image quality thresholds and the continued value of human judgment under degraded conditions.

Abstract

Automated one-to-many (1:N) face recognition is a powerful investigative tool commonly used by law enforcement agencies. In this context, potential matches resulting from automated 1:N recognition are reviewed by human examiners prior to possible use as investigative leads. While automated 1:N recognition can achieve near-perfect accuracy under ideal imaging conditions, operational scenarios may necessitate the use of surveillance imagery, which is often degraded in various quality dimensions. One important quality dimension is image resolution, typically quantified by the number of pixels on the face. The common metric for this is inter-pupillary distance (IPD), which measures the number of pixels between the pupils. Low IPD is known to degrade the accuracy of automated face recognition. However, the threshold IPD for reliability in human face recognition remains undefined. This study aims to explore the boundaries of human recognition accuracy by systematically testing accuracy across a range of IPD values. We find that at low IPDs (10px, 5px), human accuracy is at or below chance levels (50.7%, 35.9%), even as confidence in decision-making remains relatively high (77%, 70.7%). Our findings indicate that, for low IPD images, human recognition ability could be a limiting factor to overall system accuracy.

Figures (3)

• Figure 1: Ex. discarded images.
• Figure 2: Visualizing an original image with $\sim$100px IPD to low-IPD versions.
• Figure 3: Distribution of responses with respect to IPD level, with correct responses based on pair type outlined.