Spectral Reconstruction Using an Iteratively Reweighted Regulated Model from Two Illumination Camera Responses

Area of Science:

• Computer Vision
• Color Science
• Image Processing

Background:

• Accurate spectral reflectance estimation is crucial for color reproduction.
• Existing methods struggle with glare and specular highlights, affecting accuracy.
• Traditional regularized least squares (RLS) methods have limitations.

Purpose of the Study:

• To develop an improved spectral reflectance estimation method.
• To enhance accuracy in transforming RGB images to spectral reflectance.
• To mitigate the impact of glare and specular highlights in imaging.

Main Methods:

• Developed an iteratively reweighted regulated model.
• Integrated polynomial expansion signals with a cross-polarized imaging system.
• Captured two RGB images under different illumination conditions.

Main Results:

• Achieved 23.8% improved accuracy in mean CIEDE2000 color difference.
• Demonstrated 24.6% improved accuracy in RMS error compared to RLS.
• Results show sufficient accuracy within typical graphic arts industry tolerance (<3 DE units).

Conclusions:

• The proposed method significantly improves spectral reflectance estimation accuracy.
• Cross-polarized imaging effectively reduces glare, enhancing colorimetric and spectral predictions.
• The method offers a viable solution for accurate spectral property prediction in practical applications.