In land-use data general-ization, the removal of insignificant parcel with small size is the most fre-quently used operator. Traditionally for the generalization method, the small parcel is assigned completely to one of its neighbors. This study tries to improve the generalization by sepa-rating the insignificant parcel into parts around the weighted skeleton and assigning these parts to different neighbors. The distribution of the weighted skeleton depends on the com-patibility between the removed object and its neighbor, which considers not only topological relationship but also distance relationship and semantic sim-ilarity. This process is based on the Delaunay triangulation model. This paper gives the detailed geometric al-gorithms for this operation. 相似文献