Point set denoising with retained features based on adaptive Gaussian mixture model

The aim of point set noise reduction is to reconstruct noiseless point sets from noisy inputs, which is a basic problem in 3D geometry processing. The main challenge of point set noise reduction is to remove the noise while maintaining the characteristics of the underlying geometry. Existing state-of-the-art point-set noise reduction methods still have problems: some methods are not designed to recover sharp features, and some methods do not maintain geometric features well, especially detailed features. In this paper, a new method based on Adaptive Gaussian mixture model (Adaptive GMM) is proposed. By introducing the local geometric feature adaptive mechanism, the number of components, mixed weights and covariance matrix of GMM can be dynamically adjusted according to the local geometric features, so as to retain the features more effectively. With the input of the point cloud with noise and the filtered normal vector, a denoising and feature-preserving point set can be reconstructed by using the algorithm proposed in this paper. The experimental results show that the adaptive algorithm is superior to the existing methods in noise reduction quality and feature retention ability.