Geometrical feature extraction of complex curved parts is of great significance to the detection of machining quality and the reverse reconstruction of CAD model. A method for extracting feature points from scattered points of complex surface parts is proposed in this paper. Firstly

a method of neighborhood principal component analysis based on Gauss weight is proposed

estimating the degree of local change of each point neighborhood to initialize labels. Subsequently

feature points are extracted by automatic recognition based on labels. Finally

the anomaly points of the feature points are removed by the method of feature points clustering. The algorithm directly operates on scattered point clouds

without any topological connection information. The experimental results show that the algorithm is simple and effective

without too much manual adjustment parameters

and the extracted feature points are complete.