Abstract: A robust high-order bilateral denoising algorithm is introduced in the paper．By using techniques from robust statistics, awe obtain the approximations of the underlying surfaces．This leads to a conceptually good local patch with precise geometry properties, aand as a result, athe influence of noises is reduced．In addition, awe also present a diffusion method to smooth the geometry properties, a including principle curvatures and Frenet frame．The method ensures that the surfaces are smooth in the curvature space, aand maintain the details, afeatures and optimize the normals．The results show that, aeven in a high level noise situation, aour algorithm can remove noise effectively, aand at the same time keep the details of the model．More importantly, athe model may avoid shrinkage that usually comes up in the previous filtering algorithms．