3D points cloud gained by using phase-shift measurement system unavoidably has noise. Cloud-point is divided into large scale noise points and small scale noise points by density k-means clustering algorithm

isolated noise points is removed by setting neighbourhood size and quantity of points; take ratio of intraclass distance and between-class distance as evaluation function obtain the optimal number of clusters to remove little noise points cloud. Using stronger robustness improved bilateral filterer for points cloud. Experiment shows that point clouds denoising algorithm combines density k-means and improved bilateral filtering can effectively remove noises point

maintain point cloud features. Compared to mean curvature algorithm and the bilateral filtering algorithm based on feature selection

the de-noising efficiency are improved around 24 % and 16 %.