Fast reconstruction of implicit surface from large scatter point cloud with noises

This paper addresses the problem of reconstructing implicit surface from scatter point cloud with noises. We introduce a filtering operator based on improved trilateral filter, which can reduce noise effectively, while preserving sharp feature. The filtered data points are divided into subdomains using an adaptive octree subdivision method, and a local compactly supported radial basis function is constructed at each subdomains. Finally, all the local shape functions with their weights are summed up using the method of partition of unity to approximate the global domain. Numerical experiments demonstrate robust and high quality performance of the proposed method in processing a great variety of 3D reconstruction from point clouds with noises.