ABSTRACT
This chapter presents a method for automatically training structuring elements for locating specified patterns in an image. In the most robust analysis sequences, there may be large number of feature images, each generated by a unique structuring element. The simplest way to create a binary segmented image is to apply a threshold. If the background is too nonuniform, the image can be normalized through three- dimensional gray-level techniques. Rank-order operators can be defined that also operate on matrices of images. The technique of using one “golden” training image relies on the robustness of the training procedure and the several layers in the morphological network to provide good recognition under degraded conditions. Morphological networks are useful for real-time industrial applications and can be succinctly and eloquently formalized in terms of matrix morphology, where each layer in a network is defined with a matrix structuring element.
