Multilateral Filter Denoising Method Based on Meanshift for Point-Sampled Model

In order to remove the noise in point-sampled model, and prevent vertex drift and over-smoothing phenomenon, a multilateral filter denoising algorithm based on Meanshift for point-sample model is presented. First of all, the algorithm defines the neighborhood for each sampling point based on Euclidean distance; normals and curvatures of vertices in the neighborhood are estimated by covariance matrix analysis and the local optimal neighborhood of sampled point is determined by using Meanshift method. Then, the pre-smoothing normals of sampled points are calculated in the neighborhood to obtain the filter reference plane. Finally, by moving vertices to new locations, the final model after denoising is obtained. The experimental results show that our algorithm can effectively remove the noise, and can maintain local geometric characteristics of the model.