A fuzzy object retrieval system for image understanding

The authors address the problem of object recognition in the field of image understanding systems. They show how the problem can be approached exploiting information retrieval techniques. That is, given a collection of object models, the target object is regarded as a query and object recognition is regarded as a query processing task aimed at locating the models which best fulfil the given one. Since the problem is fuzzy in nature, the inference mechanism is based on a fuzzy set theoretical framework. The authors also show how the object-oriented paradigm provides the right framework to represent effectively and efficiently the knowledge about the objects. Each real world entity in the domain is mapped on a collection of composite objects, hierarchically organized, each level representing a specific abstraction level at which the entity is regarded. The authors discuss implementation issues with reference, to the retrieval component and related techniques that have been devised to improve the computational efficiency of the system.<<ETX>>