Using ensemble learning for classifying artistic styles in traditional Chinese woodcuts

Area of Science:

• Art History
• Computer Vision
• Digital Humanities

Background:

• Chinese woodcut art boasts a rich history with diverse styles and techniques.
• Existing methods lack automated approaches for identifying and classifying woodcut artistic styles and periods.

Purpose of the Study:

• To develop an automated, hybrid method for identifying and classifying Chinese woodcut artistic styles and time periods.
• To enhance the scientific and applied understanding of Chinese art through computational methods.

Main Methods:

• Image compilation, preprocessing, and color conversion for feature extraction.
• An ensemble model using Convolutional Neural Networks (CNN) for initial predictions.
• A hybrid approach combining CNN predictions with a Classification and Regression Tree (CART) meta-model for improved accuracy.

Main Results:

• The proposed hybrid method achieved significant improvements in accuracy (4.7%) and precision (4%) for Chinese woodcut classification.
• Demonstrated high levels of accuracy and precision in identifying artistic styles and time periods.
• Outperformed existing comparative methods in classification tasks.

Conclusions:

• The developed multi-stage approach offers an efficient solution for automatic Chinese woodcut identification and classification.
• This method contributes to advancements in both the scientific study and practical application of Chinese art.
• The model accurately and efficiently distinguishes between different artistic styles and historical periods in woodcut art.