The segmentation of point cloud data is an important link in the process of model reversal. The quality of segmentation affects the efficiency and accuracy of model reconstruction. The parts with complex micro-surface are composed of several small graphics side by side and cross-combined. It is difficult to simplify feature points and identify elements, which is a difficulty in point cloud data segmentation. According to the modeling characteristics of the model, the lower boundary points of the banded feature points are separated as the real feature points of the fitting feature line, and the elements belonging to the same graph are identified by the proximity of the end points of each element and the arrangement trend of the feature points near the end points. The regional growth algorithm with boundary constraints and the triangle cross product algorithm are used to segment the point clouds on the same surface. The experimental results show that this method can overcome the problems of excessive segmentation and insufficient segmentation when dealing with complex micro-surface point clouds, which lays a foundation for high-quality model reconstruction.