Spectral reflectance imaging with dual-illumination and RGB camera via regularized end-to-end learning.

Area of Science:

• Computer Vision
• Computational Imaging
• Material Science

Background:

• Spectral reflectance imaging offers illumination-independent material properties crucial for medicine and cultural preservation.
• Traditional hyperspectral imaging is expensive, while RGB methods face spectral ambiguity and accuracy limitations.
• A gap exists between high-accuracy spectral imaging and cost-effective, efficient solutions.

Purpose of the Study:

• To develop a cost-effective dual-illumination spectral reflectance imaging method using RGB inputs.
• To enhance spectral accuracy and reduce ambiguity compared to single-illumination RGB methods.
• To create a practical system for material property analysis with improved efficiency and affordability.

Main Methods:

• A regularization-guided end-to-end framework jointly optimizes illumination selection and reflectance reconstruction.
• Discriminative illumination learning is guided by multi-illumination gain priors and a physically-aware illumination modeling network.
• A lightweight reconstruction architecture incorporates adaptive illumination augmentation, hierarchical multi-path processing, and channel-wise recalibration with spectral and spatial losses.

Main Results:

• The proposed method significantly outperforms state-of-the-art techniques across multiple datasets.
• High-quality spectral reflectance reconstruction is achieved even with low-cost hardware.
• A physical prototype demonstrated reliability under hardware-induced spectral deviations and practical deployment potential.

Conclusions:

• The dual-illumination method provides a cost-effective and accurate alternative for spectral reflectance imaging.
• The developed framework effectively addresses the trade-off between accuracy and efficiency in spectral analysis.
• The system shows significant potential for practical applications in various fields requiring material property analysis.