A Parallel Topological Map for Image Segmentation

We present a new algorithm for topological feature mapping. It is compared to other known feature mapping algorithms like that proposed by Kohonen or Bertsch. The main difference to known algorithms is the inversion of the learning step. This makes it possible that a neural net can be seen as a cellular automaton. A neurone learns from its neighbours instead of teaching them the new value they have to adapt to. Equivalence of the algorithms is shown. The results of the algorithm are extremly high parallelization and flexibility in neighbourhood definition. This leads to new solutions for image segmentation problems.