Dynamic Growing Self-organizing Maps for Surface Reconstruction from Point Clouds

In order to improve the quality, rate of convergence and surface accuracy of point cloud reconstruction in the self-organizing neural network, the dynamic growing self-organizing neural networks algorithm is proposed in this paper. Firstly by the self-organizing maps algorithm, we construct the spherical triangle mesh as maps of the neural network and select the right loop numbers of the topology neighborhood. Then, we split the nodes and delete the unstable nodes to change the immobility of the network structure by training and learning of neural network for unorganized scattered point clouds. In addition, we optimize the grid to make the neural nodes and discrete points keep closer together. Finally, experiments demonstrate that this method can generate favorable results. Compared with the training characteristics of the self-organizing neural network, the algorithm can reduce the amount of calculation and improve the rate of convergence and surface accuracy of the scattered point clouds reconstruction. Especially, it is more apparent of effect for the reconstruction of a huge amount of data or the point clouds with a lot of noise.