Image analysis by morphological automata

摘要

A new approach in image processing oriented to the local study of an object is proposed: the morphological automaton. A morphological automaton is a self-adaptable window, i.e. a pattern which when placed on an image of an object, collects information (vector, matrix or disc of optical densities) necessary for its displacement function and for the objective (characterization or local transformation of the image) which has been assigned to it. The displacement and transformation procedures are controlled by decision rules which take the context into account. An automaton of the MP-class (variations in shape and position conditioned by the context) whose objective is to establish the medial axis by the displacement of discs of maximum size included in the object, has been built. This automaton has been applied to the determination of an integrated density profile and the straightening of a chromosome image. Using this approach, we have verified the efficiency of the automaton with respect to one classic method of successive thinnings.