When
Where
Title:
Abstract:
Head-mounted light field displays (LF-HMDs) have emerged as a promising approach for providing correct focus cues to mitigate the vergence-accommodation conflict (VAC) in conventional stereoscopic displays. Among different approaches to implementing LF-HMDs, integral imaging-based LF-HMDs (InI-LF-HMDs) offer a favorable balance between hardware complexity and three-dimensional (3D) rendering capability. However, systematic characterization of how view properties influence retinal image formation and visual artifacts remains limited. Current implementations are also constrained by limited view window size and location, crosstalk images, low optical efficiency, and restricted reconstructable depth range.
For visual-artifact characterization, a previously established simulation framework for light field displays is adopted and extended. Accumulated retinal point spread functions (PSFs), retinal modulation transfer functions (MTFs), and a perceptually weighted maximum side-lobe metric (PWMSL) are used to characterize artifact severity. The effects of view density, elemental-view fill factor, accommodation mismatch, reconstruction depth, and content spatial frequency are evaluated. Results show that image-splitting and duplication artifacts are governed by the overlap among neighboring elemental retinal PSFs. Higher view density generally reduces artifacts, while footprint-size variation produces a non-monotonic response due to competing diffraction and defocus effects.
To address system-level limitations, two approaches are developed to improve InI-LF-HMD performance. First, analytical relationships governing view window size and location are derived, and a time-multiplexed approach using a programmable shutter array is developed to expand the view window without sacrificing spatial resolution or view density. Second, a crosstalk-suppressed framework using programmable directional illumination is developed to control the angular emission of elemental images. By enforcing source-aperture conjugation, the emission cone is matched to the microlens numerical aperture, intrinsically suppressing crosstalk while improving optical efficiency. The approach also enables a tunable effective fill factor for balancing spatial resolution and reconstructable depth range. Experimental results validate the proposed methods.
Committee:
Dr. Hong Hua (Chair)
Dr. Yuzuru Takashima
Dr. Daewook Kim