Method of line drawing recognition based on intersection model

Proposed here is a method of drawing recognition that allows correct separation and recognition of lines, even when many similar lines form dense and complicated intersections. With previous techniques for line recognition, tracing-type methods based on such line properties as straightness are generally employed. As to junctions and intersections between graphic objects, trial-and-error methods have been studied that offer line recognition by means of repetitive generation of hypotheses based on some global constraints, and verification of these hypotheses. Such methods prove to be poor, however, at recognition of relations between lines within complicated intersection areas. In this paper, geometric patterns of line intersection areas are modeled under certain constraints, and it is made clear that these patterns may be considerably reduced in number by using border points between an intersection area, and its incoming and outgoing lines. Using this intersection model, a method is proposed to recognize line drawings, and efficiency of the method is verified by evaluation tests.