Robust colour calibration of an imaging system using a colour space transform and advanced regression modelling

Area of Science:

• Computer Vision
• Color Science
• Machine Learning

Background:

• Accurate color data conversion is crucial for reliable computer vision systems.
• Existing methods for converting device-dependent RGB to device-independent L*a*b* color spaces often introduce noticeable errors.
• The need for a robust and adaptable algorithm for precise color calibration persists.

Purpose of the Study:

• To develop a new algorithm for converting RGB to L*a*b* color data with minimal error.
• To enhance the accuracy of color calibration in computer vision applications.
• To create an adaptable algorithm for diverse computer vision systems.

Main Methods:

• Combined linear color space transformation with advanced multiple regression techniques.
• Transformed RGB components into variables that better represent the L*a*b* color space structure.
• Applied various regression models including multilinear regression, response surface, PLSR, and LASSO regression.

Main Results:

• Achieved a low initial color calibration error (ΔE(CAL) = 14.1) after linear transformation.
• Reduced color calibration error significantly with regression models, reaching as low as ΔE(CAL) = 2.1 with LASSO regression.
• The final algorithm demonstrated a marginal loss of color information (ΔE(CAL) = 2.35) on a laboratory system.
• Successfully classified ham classes with 100% accuracy, surpassing conventional colorimeters.

Conclusions:

• The developed algorithm accurately converts RGB to L*a*b* color data with minimal error (<2.2 color units).
• The algorithm is adaptable and easily recalibrable for various computer vision systems.
• Color features derived using this algorithm enable superior classification performance compared to traditional colorimeters.