Object Understanding Using Case-Based Reasoning with Fuzzy Attributed Graphs and Genetic Algorithms

抄録

To emphasize that models for recognition are described with domain knowledge and that objects of different structures or feature measurements can then be identified if they have the same name in natural language, this paper use object understanding to distinguish from the usual object recognition. To avoid the bottleneck of obtaining a complete description to the objects of the same name, case-based reasoning (CBR) is employed where several known instances (called cases) of the named object are considered as domain knowledge and a building-block method is used for reasoning. The cases and unknown objects are represented by fuzzy attributed graphs (FAG's). Using an algorithm proposed in this paper for calculating similarity between two FAG's, an object can be recognized (understood) based on the cases through recognizing its functional parts. To identify the functional parts of an unknown object, genetic algorithms (GA) are adopted for searching the best matched structures in cases.Algorithms are verified by some examples of line drawings of chair and non-chair.