A Multilevel Approach for Recognizing Characters in Engineering Drawings

摘要

In this article a multilevel method to recognize the characters in engineering drawings is proposed. It takes advantages of three different methods of character recognition: the method based on topological structure, the method based on geometrical information and the method based on invariant moment and self-learning. After the main recognition process, a "supervisor-clerk" model and contextual analysis are used to eliminate the confusion. THe experiment test shows that our multilevel method achieves good performance.