A Feature Significance-Based Method to Extract Terrain Feature Lines

Existing methods to extract terrain feature lines cannot sift through the results according to their feature strength. To solve this problem, a feature significance-based method is proposed to extract ridge and valley lines. First, Gaussian filtering and downsampling are done to the digital elevation model. Then, feature points are determined by global profile scans, and feature significance for each feature point is calculated according to height, drop height and gradient attributes. After feature extension, feature graph is constructed, and the minimum spanning tree algorithm is utilized to obtain the feature forest. In the following, branch decomposition is conducted and feature significance for each branch is figured out, on the basis of which branches are pruned. At last, final feature lines are obtained by postprocessings such as tail hook remove, location correction and smoothing. Experiments show that the proposed method can provide users with convenient and effective means to sift through the results based on the feature significance, and the extracted feature lines coincide with the real terrain. Furthermore, the proposed method has nice anti-noise capability, and can deal with flat terrain features fairly well.