Comprehensive High-Quality Three-Dimensional Display System Based on a Simplified Light-Field Image Acquisition...

Area of Science:

• Computer Vision
• 3D Display Technology
• Deep Learning

Background:

• Traditional light field acquisition can be complex and resource-intensive.
• Achieving high-resolution 3D visualization often requires sophisticated multi-camera setups.
• General environments pose challenges for accurate 3D data capture.

Purpose of the Study:

• To develop a high-quality 3D display system using a simplified light field image acquisition method.
• To reconstruct light field images of real-world objects with elevated quality in general environments.
• To enable high-resolution 3D visualization on integral imaging displays.

Main Methods:

• A simplified light field image acquisition method was employed to capture 3D information.
• A custom-trained, fully-connected deep neural network was utilized for viewpoint reconstruction.
• Instant Neural Graphics Primitives with hash encoding were trained for generating desired viewpoints.
• Pixel rearrangement of the elemental image array was performed for visualization.

Main Results:

• The proposed system successfully acquired and reconstructed high-quality light field images.
• The deep neural network generated desired viewpoints with consistent high quality.
• The system achieved high-resolution 3D visualization on an integral imaging display.
• Processing time was significantly reduced using the custom neural graphics primitives model.

Conclusions:

• The developed system provides a simple and effective approach for acquiring light field images from real objects.
• High-quality 3D visualization on integral imaging displays is achievable with the proposed method.
• The system demonstrates potential for advanced 3D display applications.