Pattern recognition and correction method for skeleton lines at small patch boundaries

Abstract

AbstractSkeleton line extraction is a key step in the dissolution and collapse operation of small patches in digital map generalization. Most parts of skeleton lines can be extracted well by existing methods, but these methods ignore skeleton line extraction at the local transition areas near the boundary, which is complex and diverse, resulting in topological inconsistency and visual noncontinuity. Our research focuses on this problem and puts forward a pattern recognition and correction method for skeleton lines at small patch boundaries. First, according to the semantic, topological, and distance characteristics of the nodes on the boundaries, the topological inconsistencies of skeleton lines at the boundaries are defined and classified into four patterns. Second, a classification and regression tree (CART) is trained to identify various patterns automatically. Finally, cubic Bezier curves are used to correct different patterns by considering the extension directions of the boundary and the skeleton line as constraints. The proposed method is validated using the National Geographic Census data in Guizhou Province, China. The results demonstrate that the recognition rate of topological inconsistency patterns can reach 100%, the topological relationships at boundaries are preserved, and the corrected skeleton lines are natural and smooth.