Recognition of Unstructured Objects by Morphological Learning
A large number of object identification methods is
suggested in the literature since the emergence of Computer Vision as
one of the major fields of research. Each of them appear to be context
sensitive and of specific applicability. Here a technique is
suggested where several gray level patterns are input as examples to a
learning algorithm which is based on morphological rules. The output
of the algorithm are a few structuring elements capable of recognizing
patterns that occur in real images encoding the intensity information
of the scene. Preliminary success of these trained set of structuring
elements in detecting the imperfections of nuclear fuel pellet surface
validates the applicability of the method. If the training set is
complete, the method is virtually sure to work in any practical
scene. Robustness with respect to edge strength, orientation and shape
variation is inherent in this technique. Computational simplicity of
this algorithm in detecting the patterns makes it attractive in
practical applications.
Key Words: recognition, detection, learning,
morphology, generalization, unstructured object, gray structuring
element.
Click here for an HTML version
