L₀ Normal Optimization and Feature Rectification for Point Cloud Denoising

For the trade-off between denoising and sharp feature preservation in point cloud denoising, a method is presented that can effectively remove noise and restore sharp features. First, a normal optimization based on L₀ minimization is carried out by combining the first and second order normal differences, and then pre-denoising is carried out according to the optimized normal direction. Then, a local dihedral frame is constructed to extract the sharp feature area points. Finally, the normal rectification is made based on the geometric information of the area near the sharp feature, and the final denoising is completed. Denoising experiments were carried out on surfaces of different structure types and noise intensity levels in hundreds of public model data. The results show that the method proposed in this paper not only removes outliers, but also restores the sharp features of the original structure to a higher degree than other mainstream methods.