Automated mold defects classification in paintings: A comparison of machine learning and rule-based techniques

Area of Science:

• Art Conservation Science
• Digital Image Analysis
• Machine Learning Applications

Background:

• Mold growth on fine art paintings is a significant threat to cultural heritage.
• Fungal contamination degrades artwork, necessitating effective detection methods.
• Current detection methods may lack precision and efficiency.

Purpose of the Study:

• To develop and evaluate a novel automated approach for detecting and categorizing mold defects in fine art paintings.
• To compare the performance of different classification techniques for mold identification.
• To enhance the preservation of valuable artworks through early and accurate mold detection.

Main Methods:

• Utilized Derivative Level Thresholding for feature extraction to identify suspicious regions in artwork images.
• Employed morphological filtering, Classification and Regression Trees (CART), and Linear Discriminant Analysis (LDA) for defect classification.
• Validated the methods using the Mold Features Dataset (MFD) and independent test images.

Main Results:

• Both classification methods (morphological filtering, machine learning) improved detection accuracy and precision over baseline.
• The CART algorithm demonstrated superior performance, increasing precision by 32%–53% while maintaining 96% accuracy.
• High accuracy was achieved even with an imbalanced dataset, highlighting the robustness of the CART model.

Conclusions:

• The proposed image analysis technique offers a precise and efficient method for identifying mold defects on paintings.
• The CART algorithm shows significant potential for enhancing the automated detection of mold in art conservation.
• Early and accurate detection facilitated by this method is crucial for safeguarding invaluable cultural heritage artworks.