Detecting stylistic categories and uncovering the topic characteristics of ancient murals across different styles are essential for advancing mural studies and cultural understanding. To this end, we first develop a mural style detection framework that integrates multiple visual feature fusion with logistic regression (MV2FLR). The model extracts multi-view visual features from mural images and trains a logistic regression classifier to achieve efficient and reliable style detection. Meanwhile, we further analyze the key discriminative features that differentiate stylistic categories. Subsequently, we employ the BERTopic model to perform topic mining on murals of varying styles, thereby revealing their topic distribution patterns and intrinsic connections. Experimental results show that MV2FLR achieves a precision rate of 0.9889, substantially outperforming existing baselines. Murals of different styles exhibit distinct topic distributions but also share core themes, reflecting the continuity of religious visual traditions and the visual integration of diverse belief systems.
Sun et al. (Fri,) studied this question.