Abstract: This work presents an information theoretic approach tothe problem of building extraction in aerial imagery.Boundary curvature is selected as a suitable measure ofshape information and its relationship to Shannon'sconcept of entropy is discussed. A novel concept ofglobal and local shape entropy is defined. Amathematically tractable approach to connected regiongrowing is developed utilizing this novel concept ofglobal shape entropy. An algorithm for the automaticextraction of building silhouettes in aerial imagery ispresented. The algorithm is shown to be shapeindependent by successfully extracting rectangularbuildings and a circular storage tank in real imagery.The limitations of the algorithm are discussed andpossible future research is proposed.!14
展开▼