Research of the structural-learning-based symbol recognition mechanism for engineering drawings

Considering the diversity of symbol representation in the engineering drawings and the adaptability of the symbol recognition method, we propose the structural description and learning method of symbols and implement a real drawing recognition system embedding the self- learning mechanism. In the system, the symbol templates are automatically created by the incremental sample learning and used in the recognition. Meanwhile, different types of recognition result are used to guide the further learning work: template creating or updating. Experiments show that the recognition ability will improve incrementally without modifying the program and only a few samples are needed.