We are developing a drawing image database system that stores plant diagrams. The system scans, recognizes and stores plant diagrams. Users can then refer to any parts of the diagrams according to their needs.
Similarity retrieval, a retrieval where users show sample diagrams to describe what they want, is realized by defining a similarity measure between drawing images. The measure is based on graph representations of drawing images and it is effective for retrieval of drawing images that describe logical meaning by their structure.
In this paper, a fast retrieval method that features two-level matching between graph representations is proposed. By combining rough matching and detailed matching, the average response time is reduced to 2.8 seconds, which is about one sixth of the former method. It is confirmed that the response time is adequate for applications at plant control and maintenance.
J-STAGEがリニューアルされました! https://www.jstage.jst.go.jp/browse/-char/ja/
