An Algorithm of Extracting Features from Point Set Models

Our feature extraction is a multi-step refinement method, we use principle curvatures to flag potential feature points and develop a new approach to detect potential feature curves by employing our improved weight sensitive moving least squares. Then the potential feature points are enhanced by projecting onto the local potential feature curves. Using an optimized principal covariance analysis approach, we smooth the projected points. Finally smooth feature curves are obtained after resolving gaps and relaxing the results.