Sharp Feature Preserved Filtering Of Point Cloud Based On Joint Bilateral Filters

In recent years,due to stimulus coming from increasing of virtual/augmented reality appli-cations,point cloud processing again enters into the spot of computer graphics research.This paper investigates feature-preserving filtering of point clouds.In particular,two issues,sharp feature preserved denoising and small feature removing of point clouds,are investigated,and two algorithms are respectively put forward to address the issues.Noticing that existing methods for point cloud denoising are prone to blur sharp edge fea-tures while removing noise,we propose a guided normal filter for point cloud denoising.By assigning feature points with multiple nomrals,we can establish a piecewise smooth normal field for a point cloud,which lays the foundation for recovering sharp features.In addition,we extract sharp feature curves with the aid of the L₁-median skeleton of feature candidates,which avoid the issue of non-uniform distribution in the step of point position updating.Geometric features represent rich details of point cloud models,whose scale is much larger than noise.Traditional point cloud denoising methods seem incapable of processing and ana-lyzing these features.Noticing that the rolling guidance filter can suppress small-scale features while preserving large-scale structures,we adapt it to filter the normal field of point clouds for removing small geometric features from these 3D models.We also propose a new optimization energy for updating point positions.In the proposed formulation,a weighted point constrain-t is introduced to keep non-feature points from shrinking while allow feature points moving drastically.We also employ a multi-normal strategy to recover sharp features for noisy point clouds.In order to validate our methods,we run our algorithms on a variety of point cloud mod-els.Experiments demonstrate that our algorithms can preserving sharp features well while de-noising or removing small-scale features.We also numerically depicts the convergence of our algorithms.