Voronoi Pyramids Controlled by Hopfield Neural Networks

We present an algorithm for image segmentation with irregular pyramids. Instead of starting with the original pixel grid, we rst apply some adaptive Voronoi tesselation to the image. This provides the advantage that the number of cells in the bottom level of the pyramid is already reduced as compared to the number of pixels of the original image. Furthermore the Voronoi diagram is a powerful tool for shape description and image compression. For the construction of the irregular pyramid we present a Hoppeld neural network which controls the decimation process. In this paper we extend our previous results by proving a more general theorem than in 4]. The contributions of this paper are the initialisation of the pyramid by a Delaunay graph and the extension of the results for Hoppeld neural networks for decimation. The validity of our approach is demonstrated by several examples.