Integrating AI and 3D Digital Technologies to Assist in the Stylistic Analysis of Taiwanese Artisan Paintings: A Case Study on Door God Imagery

Abstract. In Taiwanese temple decorative arts, door god paintings serve not only religious purposes but also reflect the distinct styles of individual artisans. With the advancement of digital preservation, Artificial Intelligence (AI) offers new opportunities for identifying and preserving traditional stylistic lineages. This study investigates Pan-style door god paintings, a prominent artistic lineage, and proposes a classification framework combining image feature extraction and machine learning.Fuzzy Color and Texture Histogram (FCTH) is used as the primary image descriptor. Models are trained and validated using the WEKA platform with two supervised algorithms: J48 decision tree and Random Forest. J48 offers interpretable classification logic, while Random Forest improves overall stability and accuracy through ensemble voting.Experimental results demonstrate that Random Forest outperforms J48, achieving an average accuracy of 91.2% versus 85.6%. Feature importance analysis highlights edge texture and color saturation—particularly in red and gold—as key indicators of Pan-style characteristics, which typically feature rich coloration and symmetrical composition.This study confirms that AI-driven classifiers are effective for distinguishing Pan-style aesthetics in temple paintings and can support restoration decision-making. The methodology offers a scalable model for polychrome heritage classification and preservation, adaptable to broader cultural heritage contexts.