Perception-based approach in recognition of structured patterns

This paper presents an approach to the problem of structured pattern recognition. The approach is built around the multilayered learning method and approximated reasoning techniques, which aim to convey the facilities of human perception and reasoning into the pattern recognition task. We further show how the complex ontology of an external human expert on the pattern domain knowledge can be effectively elicited and incorporated into the learning process through an interactive procedure which essentially implements the Computing with Words paradigm.