No abstract

The problems of multiple interpretations and feature interactions that occur in attributed adjacency (AA)-based feature extraction systems result from a lack of robustness in the recognition algorithm when deviations from the feature definitions occur. This stems from the fact that individual graphs are stored in the feature taxonomy and once multiple graphs interact the interaction issue occurs. In this paper, a new method is presented for defining features based on a type of hint-based taxonomy, which is rare in boundary representation schemes. This new method still uses the traditional AA graph and matrix to define the part but does not extract subgraphs. It is shown that identifying only one face and then proceeding can find a feature. The modified attributed adjacency (MAA) scheme is used to define the part which allows more information to be stored in the part representation (graph or matrix), and this allows multiple interpretations to be solved.

No abstract