Inland-usedatageneralization,theremovalofinsignificantparcelwithsmallsizeisthemostfrequentlyusedoperator.Traditionallyforthegeneralizationmethod,thesmallparcelisassignedcompletelytooneofitsneighbors.Thisstudytriestoimprovethegeneralizationbyseparatingtheinsignificantparcelintopartsaroundtheweightedskeletonandassigningthesepartstodifferentneighbors.Thedistributionoftheweightedskeletondependsonthecompatibilitybetweentheremovedobjectanditsneighbor,whichconsidersnotonlytopologicalrelationshipbutalsodistancerelationshipandsemanticsimilarity.ThisprocessisbasedontheDelaunaytriangulationmodel.Thispapergivesthedetailedgeometricalgorithmsforthisoperation.
