RealSLF and FlexiDim: towards practical spectral light field imaging.

Area of Science:

• Optics and Photonics
• Computer Vision
• Machine Learning

Background:

• Spectral light field (SLF) imaging captures rich angular and spectral data, surpassing conventional RGB imaging for scientific and industrial uses.
• Existing SLF research is hindered by a lack of high-resolution real-world datasets and efficient, resource-aware learning frameworks.

Purpose of the Study:

• To introduce RealSLF, a comprehensive real-world SLF dataset with high spatial, angular, and spectral resolutions.
• To develop FlexiDim, a flexible and computationally efficient learning framework for high-resolution SLF reconstruction.
• To demonstrate a practical, scalable RGB light field system for SLF data acquisition.

Main Methods:

• Created RealSLF, a large-scale dataset with diverse spectral bands and dense viewpoints, nine times larger than existing datasets.
• Implemented a geometric correction pipeline to improve cross-view consistency and reduce misalignment.
• Proposed FlexiDim, a resource-affordable learning framework integrated with super-resolution modules for SLF reconstruction.
• Developed a prototype RGB light field system using a microlens array for practical SLF data capture.

Main Results:

• RealSLF provides an unprecedented scale and resolution for SLF research.
• FlexiDim demonstrates effective high-resolution SLF reconstruction under computational constraints.
• The prototype system validates the feasibility of scalable, component-free SLF imaging.

Conclusions:

• The RealSLF dataset and FlexiDim framework significantly advance SLF imaging capabilities.
• These contributions address critical limitations in data availability and computational efficiency for SLF applications.
• The developed system offers a practical and scalable solution for capturing high-resolution spectral light field data.