MST Clustering Of Street Buildings Based On The Gestalt Principles

Building clustering aims to divide buildings into different groups according to the principles and laws of human visual perception through spatial clustering methods,so that the similarity of buildings within the same group is as large as possible.Researching clustering methods of buildings has important implications for the multi-scale expression of spatial objects,the identification of geographic element distribution rules,and the comparison of spatial scene similarities.Therefore,this paper uses the Minimum Spanning Tree(MST)as the basic clustering algorithm and the Gestalt criterion of building pattern recognition to cluster the street buildings,aiming to propose a spatial clustering method that is more in line with human visual cognition and is an effective means to detect spatial distribution patterns,it provides strong support for the comparison of spatial scene similarities.In this paper,the principle of Gestalt psychology,which is applicable to the distribution pattern of buildings,is studied first,and the calculation methods of various Gestalt factors describing the geometry of buildings are discussed.After that,the geometric model used in building clustering— Constrained Delaunay Triangulation(CDT)is described and optimized.Based on the CDT,the neighbour-graph describing the adjacency relationship between buildings is obtained,and the adjacent distance between buildings is calculated.Based on the “visual distance” proposed by Ai(2007)that combines the Euclidean distance,direction difference,and size difference,the Gestalt factors considered in the “visual distance” is supplemented and improved.The proximity distance is used to describe the distance between buildings,shape differences and density differences are added,and the height of the building is also used as a constraint for building clustering.At last,the MST is constructed with the visual distance of the building as the weight,and the clustering result of the street building with the visual distance constraint is obtained through the MST pruning.In order to verify the validity of the MST clustering results of buildings with constraints of visual distance,different constraints and different clustering methods in the same experimental area were compared and analyzed.Firstly,MST clusters are constructed with three different distances as constraints,including the Euclidean distance between the centre point of the building,the average distance between visible scope of buildings,and the visual distance consists of the Gestalt factor and the proximity distance.Secondly,under the constraint of visual distance,building clustering are performed with three different clustering methods,including MST,K-Means,and hierarchical clustering.The experimental results show: on one hand,in the clustering results based on visual distance constraints,the coefficient of variation(CV)of the geometric parameters of the buildings within each sub-cluster is smaller than that of the other two distance constraints,indicating that the clustering results obtained by the constraint of visual distance are significantly better than the results of the other two distance constraints.On the other hand,in the clustering results obtained by MST method,the CV of the geometric parameters of the buildings within each sub-cluster is smaller than the clustering results of the other two clustering methods,indicating that the results obtaned by the MST method are superior to the other two clustering methods.Therefore,it can be considered that the results obtained by MST method,which considering the visual distance of the Gestalt principle as a constraint are reasonable,and the effect is in line with human visual perception.