Colorimetric Characterization of Color Imaging System Based on Kernel Partial Least Squares

Area of Science:

• Color Science
• Image Processing
• Machine Learning

Background:

• Accurate colorimetric characterization is crucial for color information management in digital imaging.
• Existing methods like nonlinear regression and neural networks have limitations in precision.

Purpose of the Study:

• To propose and validate a novel colorimetric characterization method using kernel partial least squares (KPLS).
• To improve the prediction accuracy of color space transformations in imaging systems.

Main Methods:

• Utilized kernel function expansion of RGB response values as input features.
• Employed CIE-1931 XYZ as output vectors for the KPLS model.
• Determined hyperparameters via nested cross-validation and grid search.

Main Results:

• The KPLS model achieved superior performance compared to weighted nonlinear regression and neural network models.
• Experimental validation using ColorChecker SG chart demonstrated high prediction accuracy.
• Evaluated using CIELAB, CIELUV, and CIEDE2000 color difference metrics.

Conclusions:

• The proposed KPLS method offers a robust and accurate approach for colorimetric characterization.
• This method significantly enhances color information management in color imaging systems.
• The KPLS model provides a reliable color space transformation with good prediction accuracy.