Three-dimensional (3D) semantic segmentation of Buddhist heads is essential for digital heritage applications such as virtual restoration, conservation, and art-historical analysis. However, existing segmentation methods face challenges due to the complex geometry and degraded surfaces of heritage statues. This paper proposes an efficient method for face segmentation from 3D point clouds of Buddhist heads, leveraging geometric features and topological relationships. The approach comprises three stages: (1) symmetry-based rotation for orientation normalization, (2) vertical gridding and color mapping for depth-aware parameterization, and (3) hybrid segmentation using a Face Topology Graph (FTG) and the Segment Anything Model (SAM) with point and box prompts. Experiments on 50 Buddhist heads demonstrate the method’s efficiency and robustness, achieving an average IoU of 0.73 across seven facial components. The proposed workflow provides a scalable solution for semantic 3D modeling of heritage artifacts, supporting accurate analysis and interactive visualization.