Spectral imaging using consumer-level devices and kernel-based regression.

Area of Science:

• Computational imaging
• Color science
• Spectroscopy

Background:

• Hyperspectral imaging typically requires specialized equipment.
• Consumer-level devices offer potential for cost-effective imaging solutions.
• Accurate spectral reflectance estimation is crucial for material analysis and digital archiving.

Purpose of the Study:

• To evaluate the feasibility of hyperspectral reflectance factor estimation using a consumer laptop display as an adjustable light source.
• To assess the impact of training set size and composition on estimation accuracy.
• To compare different kernel-based regression models for spectral estimation.

Main Methods:

• Hyperspectral reflectance factor estimation in the 400-700 nm range using a trichromatic camera (Nikon D80) and a laptop display (Dell Vostro 2520).
• Utilized targets included ColorChecker Classic, Munsell Matte samples, tempera icon paintings, and human hands.
• Employed kernel-based regression models (polynomial and Matérn kernels) with simplified training sets and evaluated model optimization techniques.

Main Results:

• Demonstrated accurate spectral and color estimations, with significant improvements in DE00 color distance (up to 99%) and Pearson distance (up to 99%).
• Showcased that modest, representative training data (e.g., Digital ColorChecker SG) markedly enhance estimation accuracy.
• Achieved high accuracy for challenging targets like paintings and human hands even with simplified models.

Conclusions:

• Consumer-grade equipment, specifically a laptop display and a standard camera, can achieve reliable hyperspectral reflectance estimations.
• The study highlights the effectiveness of small, representative training datasets in optimizing estimation models.
• This approach offers a simplified and accessible method for spectral imaging applications.