Color Conversion of Wide-Color-Gamut Cameras Using Optimal Training Groups

Area of Science:

• Color Science
• Computer Vision
• Display Technology

Background:

• Accurate colorimetric conversion is crucial for wide-color-gamut (WCG) displays.
• Establishing precise conversion models for WCG cameras presents significant challenges.
• Existing methods struggle to achieve desired approximation accuracy across the entire color space.

Purpose of the Study:

• To propose an optimal method for establishing color conversion models from camera RGB to CIEXYZ space.
• To enhance the accuracy of color conversion for wide-color-gamut applications.
• To investigate methods for improving color conversion model performance.

Main Methods:

• Utilizing the Pearson correlation coefficient to assess linear correlation between RGB and XYZ values.
• Selecting optimal training data groups based on high linear correlation.
• Developing and testing color conversion models, including polynomial transforms and BP artificial neural networks (BP-ANN).
• Analyzing sample groups divided by hue angles and chromas in CIE1976L*a*b* space.

Main Results:

• The proposed optimal method effectively identifies training groups with superior linear correlation.
• Color conversion models trained with optimally selected groups demonstrate improved accuracy.
• Experimental results show reduced color conversion errors (CIE1976L*a*b* color difference) when using hue-angle-divided training groups for polynomial transforms.
• BP-ANN models also showed performance, but polynomial transforms with hue-based grouping were particularly effective.

Conclusions:

• An optimal method using Pearson correlation significantly improves the accuracy of RGB to XYZ color conversion models for WCG cameras.
• Dividing training data by hue angles enhances the efficiency and accuracy of polynomial-based color conversion models.
• The findings contribute to more accurate color reproduction in wide-color-gamut display systems.