Segmentation and association among lines and junctions for a line image

Abstract The relationships among line primitives and junctions for a line image are of great importance in pattern recognition. With a conventional thinning method, the thinning result may yield several short branches and multiple fork points at a junction, which is the intersection of line primitives, from the view of an original line image. This is called the lability of the shape of the skeleton and will become an impediment in the construction of relationships among line primitives and junctions. In this paper, an effective approach to performing the segmentation and association among line primitives and junctions for a line image is presented. The proposed approach cannot only segment the line primitives and junctions but also construct their relationships for the further interpretation of the given line image. The segmentation is first performed by the analysis of line continuation to construct a map of uni-OLLS (orientation of the longest line in sight) and a map of multi-OLLS for an input line image. The former is used to find the line primitives, the latter to find the junctions. Next, the two orientational maps are processed individually via a 2D median filter, 2D labeling and partitioning, and symmetry measuring. Then the line primitive containing the orientation and symmetry information, and the junction containing the reasonable central point information are obtained. Finally, the relationships among the found features are constructed by the transition of the orientation from the partitioned line primitives to the junctions. The experimental results show that the proposed approach is feasible for the line type of characters under consideration.