Segmenting surfaces: a comparison between the performance of a neural tree and a back-propagation algorithm

A neural tree (NT) is a tree of neural networks. The term neural network (NN) is frequently used for indicating a class of algorithms which take advantage of a set of distributed elementary processing units for computing. We use the processing structure called linear machine (LM), a generalisation of perceptron. A tree of LMs is constructed, and each node of the tree (a LM) is instructed to recognise fragments of surfaces according to the classification of differential geometry. A comparison between the performance of a NT and a single backpropagation (BP) algorithm for classifying surface fragments is presented