Recognition of Handwritten Mechanical Drawing by Neural Network. 1st Report. Recognition Efficiency of NN for Line Extraction.
Zhen WANG, Kouji Tsumura, Yoshio Saito · TRANSACTIONS OF THE JAPAN SOCIETY OF MECHANICAL ENGINEERS Series C · 1996
In order to satisfy the requirement of autodrawing-data-in in a CAD/CAM system, the research on hand written mechanical drawing recognition is becoming more and more important. The key step in the process is the extraction and separation of drawing features, such as lines and intersections. In this paper, we present a new method which uses a neural network approach for extracting these features. We investigated the input and output characteristics of BP algorithm, and elucidated the main factors controlling the effect of learning efficiency and recognition ratio. They include the humming distances among learning patterns and the ratio of active portion to passive portion in the input layrer. Summarizing the above results, we developed a method that determines the optimal structure of the neural network. Several examples have shown the success of the NN system in recognizing mechanical drawings under many difficult conditions.