Optimal decomposition of large-neighborhood CNN templates

From a practical point of view, template decomposition is an important issue since image processing operations can be non−local while the cells in physical CNN realizations often only communicate with cells in their direct neighborhood. In this paper the existing relation between Mathematical Morphology and CNNs is used to derive an optimal template decomposition strategy for a specific class of templates. The strategy is based on the decomposition of morphological structuring elements.