Subjective Quality Assessment of Foveated Omnidirectional Images in Virtual Reality

Abstract

This study presents a novel dataset called “Foveated Omnidirectional Image Quality Assessment” (FOIQA) for the subjective quality evaluation of foveated 2D omnidirectional images. This dataset addresses the limitations of existing datasets by leveraging a high-resolution head-mounted display and a gaze-contingent evaluation approach. We provide individual opinion scores, mean opinion scores, and gaze data associated with both the test and reference images. The utility of our dataset is validated by benchmarking two existing objective foveated image quality metrics. Our results demonstrate that incorporating gaze data into the evaluation framework improves the accuracy of one of the tested objective metrics. The dataset is publicly available at https://doi.org/10.5281/zenodo.14009106.