Freeform surface reconstruction by RBF neural network

A novel method for surface reconstruction by radial basis function (RBF) neural network is proposed. An appropriate RBFNN model is constructed. The relationship between basis function and property of constructed surface is discussed. Theory analysis and the experiment results show that the surface constructed by some commonly used basis function is smooth. The precision and the training speed of RBF for surface reconstruction are superior to that of MLP. This approach is valuable for practical application.