Linear building pattern recognition combining Gestalt principles and convex polygon decomposition

Building patterns are important local structures characterizing urban areas. Building patterns in previous studies are mostly recognized based on the Gestalt principles in which buildings are considered as a whole. However, human vision is also proved as a parts-based system, and some visually aware patterns may fail to be recognized with the existing methods. This paper first combines Gestalt principles and the convex polygon decomposition to recognize linear patterns. First, the linear patterns are defined based on the triples and Gestalt principles. Second, linear patterns are recognized combining the convex polygon decomposition, and the buildings' orthogonal features are considered in their decomposition. The experimental results show that proposed method is effective to recognize the linear patterns in study area. Compared with the existing methods, the accuracy and recall have increased by 15.7% and 30.5%, respectively.