Reading Between the Eyes: Assessing the Impact of Interpupillary Distance on Face Recognition

Interpupillary distance (IPD) is a crucial measure of image quality in face recognition, representing the pixel distance between a subject's pupils. Many recognition systems require a minimum IPD for reliable matching, yet operational settings often provide low-IPD images that require upscaling. This work examines the impact of IPD variation and upscaling choices on recognition performance using two state-of-the-art deep learning models. We generate IPD variants (5px-30px) from a subset of the MORPHv3 dataset and apply multiple interpolation techniques to evaluate resolution enhancement. Image quality and semantic retention are further analyzed using image quality assessment (IQA) metrics. Results show that lower IPDs significantly degrades recognition accuracy, with some models exhibiting greater sensitivity to resolution loss. While interpolation methods help preserve visual details, they fail to fully recover biometric features, limiting recognition reliability. Our findings underscore the shortcomings of traditional upscaling and highlight the need for adaptive preprocessing strategies and resolution-aware models to enhance performance in real-world applications, including forensic investigations and surveillance.