Output details

Parameterization of point-cloud freeform surfaces using adaptive sequential learning RBF networks

<23>Collaborative research using expertise in machine learning and computer graphics, reporting novel use of RBF neural networks for parameterization and reconstruction of freeform surfaces from range images. An adaptive learning algorithm allows neurons to be dynamically inserted and fully adjusted according to underlying data. Extensive testing on public datasets shows the RBF network is a simple, low-cost, low-storage solution to many practical problems in 3D modelling, e.g. construction, re-sampling, hole-filling, multiple level-of-details and data compression. Theoretically, the adaptive learning contributes as a generic solution for effectively learning distribution features, thus enabling functional representation of large unstructured datasets.