An octree-based point cloud compression algorithm is proposed from 3D point cloud data that are without any foreknowledge of information. The algorithm improves the stop condition of segmentation to stop dividing at the right depth and to ensure appropriate voxel size; and in the data structure, each node is assigned a bitmask; by manipulating bitmask, query and remove data when traversing; later position encode the points and thus effectively remove the outliers and surface noise, increase point cloud compression efficiency with range encoding. Experimental results show that the algorithm much completely retains key information of point cloud, and has some theoretical significance and application value for point cloud data preprocessing.