Empirical Evaluation of Virtual Fonts with Enhanced Out-of-Focus Legibility in the Presence of Inaccurate or Incomplete Eye Accommodation Demand

Abstract

One of the key challenges in optical see-through (OST) augmented reality (AR) environments is accommodation mismatch caused by the fixed focal plane of commercial head-mounted displays (HMDs). Devices such as the Microsoft HoloLens 2 render all virtual content at a fixed distance of two meters. So, retinal blur occurs when users attempt to view virtual content rendered at different distances. This blur significantly impairs text legibility and can lead to serious consequences in fields where accurate information transmission is critical, such as healthcare, manufacturing, and navigation. SharpView addresses this issue through a pre-compensation method that combines Zernike polynomial-based blur modeling with Total Variation (TV) optimization, ensuring text remains sharp even under blurring conditions. This thesis further improved the algorithm in two respects. First, the fourth-order Zernike term was incorporated into the OTF generation process to model the spherical aberration of the human eye more accurately. Second, a white masking technique was introduced alongside the existing black masking method to eliminate experimental bias caused by brightness differences between SharpView images and the Control. Using this improved algorithm and the Microsoft HoloLens 2, a commercial OST AR device, a human subject experiment with 16 participants was conducted under two accommodation mismatch conditions. Incomplete Accommodation refers to a condition where virtual content appears before the eyes have fully focused on the new distance, while Incorrect Accommodation refers to a condition where the eyes focus on the display’s fixed focal plane rather than the actual rendering distance of the virtual content, resulting in persistent blurriness. The experiment considered three letters (A, E, and O), three levels of CL (%) (70%, 80%, 90%), and two levels of τ (high, low). Two focal disparities were considered (3.5Diopters and 4.5Diopters). The experimental results showed an overall SharpView selection rate of 32.1% in Experiment 1 (Incomplete Accommodation) and 22.6% in Experiment 2 (Incorrect Accommodation), with participants generally preferring the Control. This tendency is consistent with the mere-exposure effect, in which familiarity with a stimulus increases preference independent of its objective quality. Nevertheless, under certain parameter combinations,specifically CL levels of 70% and 80%, High τ, and the letters A and E, SharpView’s selection rate exceeded chance level, although these differences did not reach statistical significance after Bonferroni correction. High τ consistently increased the SharpView selection rate across both aberration conditions and both experiments, whereas a CL of 90% consistently reduced it due to excessive ringing. SharpView was more effective for letters composed of straight or diagonal lines (A, E) than for the curved letter O, although this pattern was not confirmed under the 4.5D condition of Experiment 1. These findings indicate that SharpView, in its current form, did not clearly outperform the Control in either experiment. However, by identifying the conditions under which SharpView is ineffective, this study provides practical parameter guidelines for improving text legibility in future OST AR systems.