Unsupervised learning of 3D objects conserving global topological order

A model is proposed for global representation and cognition of 3D objects, and a global self-organizing learning method using Kohonen´s local topology conserving maps is presented. A 3D object is represented by a set of sheets (subnets). Each of the sheets is mapped onto a part of the surface of the subject by a locally converging Kohonen map. The adjacency of the overlapped sheets is also defined over the whole surface of the objects. Thus, these adjoining subsets will eventually converge to a global atlas representation of any closed surface (as a combinatorial manifold), the Kohonen maps on each part become the local coordinate (or chart) maps. The proposed model is able to represent the objects distributively and can easily accommodate local features