Engineering drawings on paper have several disadvantages; they require a large storage space, are subject to deterioration, and are difficult to reuse. Several systems have been proposed to convert drawings to computer data. Most of them, however, rely on simple bottom-up algorithms and are very time-consuming because they must depend on human intervention for resolving recognition bottlenecks. Here, an automatic recognition system of mechanical engineering drawings for this paper-CAD conversion is proposed. It utilizes an enhanced bottom-up scheme for component recognition and a top-down scheme to identify valid component structures. Processing units are realized as individual experts and are integrated using a blackboard data system. This structure makes the proposed system robust to both noise and variations of drawings format and allows parallel implementation. Experimental results show that the system works well on the drawings with connected elements such as character strings and dimension lines.
