Spectral Reflectance Reconstruction Using Fuzzy Logic System Training: Color Science Application.

Area of Science:

• Color Science
• Machine Learning
• Fuzzy Logic Systems

Background:

• Spectral reflectance recovery is crucial for accurate color reproduction.
• Existing methods often lack interpretability or are outperformed by newer techniques.

Purpose of the Study:

• To apply fuzzy logic for the first time to spectral reflectance recovery from CIEXYZ and RGB values.
• To develop an interpretable machine learning model that outperforms classical methods.
• To extract and understand the learned rules within the fuzzy system.

Main Methods:

• Training a fuzzy logic inference system (FIS) using the Macbeth ColorChecker DC dataset.
• Testing the FIS performance on a dataset of 130 artist's paint samples.
• Comparing the FIS performance against established spectral recovery methods.
• Extracting and analyzing the fuzzy rules learned by the system.

Main Results:

• The developed FIS accurately recovers spectral reflectance.
• The fuzzy logic approach significantly outperforms previous methods both spectrally and colorimetrically.
• The system's learned rules provide interpretable insights into the relationship between RGB/XYZ inputs and spectral outputs.
• The system utilizes four reference spectral curves combined non-linearly.

Conclusions:

• Fuzzy logic offers a powerful and interpretable alternative for spectral reflectance recovery.
• The extracted rules from the FIS provide valuable knowledge, unlike 'black box' models.
• This approach can be extended to other problems in color and spectral science, offering a pathway for knowledge discovery.