Indoor scene plays an important role in human living environment which constitutes the urban landscape together with building and flora reconstruction. However, the accessible 3D indoor scene data are usually not as semantic and reasonable as expected. They may contain a multitude of topological errors and omissions. This easily results in fragments production during model reusing and incompatible transform during format conversion. We creatively build a scene hierarchy using “Graph Cuts” method iteratively by considering the topological relation (“support & non-support” and “primary & non-primary”) and geometry relation (symmetry, alignment, intersection and connectivity). The hierarchy reorganizes the scene in different levels and users could select the required level of detail which contains semantic segmentation. It corrects the topological errors and offer convenience for further scene edit.