A Vector Building Clustering Algorithm Based on Local Outlier Factor

Objectives The mining of building cluster features in topographic maps is of great significance to realize automatic cartographic synthesis and spatial knowledge mining, but it is difficult to identify building polygon communities with different distribution densities and morphological characteristics in cities.

Methods A clustering method combining local outlier factor (LOF) is proposed. Based on the adjacency map of buildings, feature vectors are constructed according to the differences of form factors and proximity distances between adjacent buildings. LOF algorithm is used to dynamically calculate the anomaly degree of feature vectors and eliminate the abnormal vectors. And the building cluster is obtained.

Results By adjusting the upper limit of LOF local anomaly factor and proximity number, the final clustering results are obtained. The experimental results show that the proposed method can effectively identify and distinguish densely distributed building groups in the city.

Conclusions We provide a new way to solve the clustering problem of urban dense buildings, verify the importance of Gestalt criteria, and the clustering results are consistent with human visual cognition.