The theory of morphological neural networks and their applications have witnessed a continuous growth in the last few years. In this context, calculating the next state of a neuron or performing the calculationof the next layer involves one of the four elementary operations of mathematical morphology:erosion, dilation, anti-erosion, and anti-dilation. In its most general form, mathematical morphology can be conducted in the context of complete lattices. Morphological associative memories are among the different morphological models that are being developped. We believe that it will be possible to introduce improved versions and new applications for these modified models. Specifically, we intend to develop morphological models based on ideas that are drawn from fuzzy set theory and regulated mathematical morphology in order to increase the tolerance with respect to noise and to eliminate spurious memories. The new models will be tested in experiments with problems in computer vision.
News published in Agência FAPESP Newsletter about the scholarship: