Quantifying the High Tang Aesthetic through Geometric and Computational Analysis of Buddhist Sculpture

Chronological assessment of rock-cut cave heritage often relies on qualitative typological analysis, which remains subjective and lacks metric standardization. This study proposes a quantitative interdisciplinary framework for dating Buddhist sculpture by integrating high-precision 3D laser scanning with Explainable Artificial Intelligence (XAI). Focusing on Tang Dynasty statues from Guangyuan (7th–8th century CE), we analyzed fourteen carefully selected sculptures using a small-sample, high-dimensional strategy. Standardized orthographic projections derived from high-fidelity 3D models enabled extraction of sixteen geometric features. Results reveal a measurable morphological transition: Phase III statues exhibit significant mid-face vertical elongation and reduced workshop standardization, while Phase IV works display a “square-round” morphology driven by vertical facial compression, reflecting High Tang humanistic aesthetics. Machine-learning models, particularly Support Vector Machines and Logistic Regression, achieved 92% chronological classification accuracy. SHAP analysis identifies eye height and eye-to-nose distance as primary determinants of stylistic change, demonstrating a scalable pipeline for computational heritage dating.