High-quality integral imaging 3D display from a captured monocular image.

Area of Science:

• Optics and Photonics
• Computer Vision and Image Processing

Background:

• Three-dimensional (3D) integral imaging (InI) faces limitations in spatial resolution and viewpoint density due to space-bandwidth product constraints.
• Generating 3D light fields for real scenes is hindered by the cost and complexity of traditional light field sampling methods.

Purpose of the Study:

• To propose a high-quality light field rendering pipeline that overcomes the limitations of conventional InI systems.
• To enable high-quality 3D display using a single captured monocular image, reducing hardware complexity.

Main Methods:

• A depth generator (DG) predicts depth maps from monocular images.
• Forward light tracing (FLT) synthesizes elemental image arrays (EIAs) using monocular images and depth maps.
• An elemental image super-resolver (EISR) enhances the spatial resolution of the generated EIAs.

Main Results:

• The proposed pipeline significantly improves depth map prediction accuracy (54% increase in structure similarity).
• Reconstructed 3D images show a 50% improvement in peak signal-to-noise ratio compared to conventional methods.
• Optical experiments validated high-quality 3D display for both virtual and real scenes with a streamlined sampling process.

Conclusions:

• The developed light field rendering pipeline effectively enhances 3D integral imaging display quality.
• This approach facilitates the miniaturization and integration of light field sampling devices, making advanced 3D displays more accessible.
• The method offers a cost-effective and efficient solution for generating high-fidelity 3D light fields from monocular inputs.