| 摘要 |
<p>A technique for recognizing hand-drawn graphic symbols operates on vectorized graphic data, including lines, arcs, and circles, that have been produced by an automatic drawing recognition system. The user creates segmented rule bases for recognizing various symbols by describing the geometric structures of recognizable symbols and subsymbols. The rule bases each are compiled into a respective decision tree. Fetched input vectors and entities are placed in software "blackboards", and more particularly, into appropriate symbol slots and subsymbol slots thereof. The growing symbol and subsymbols on the blackboards are continually restructured until they form closed loops or the blackboards are filled. The decision tree then is applied to all possible permutations of the symbols and/or subsymbols to accomplish partial or complete recognition. An inference engine subsystem controls application of the decision trees to limited bits of the symbols and subsymbols, matching decision subtree segment labels to attributes associated with the blackboard data, and orders the matched data according to degree of matching to effectuate application of the decision tree to the linked lists.</p> |
