The Gestalt Principle for Building Area Extraction

Building area extraction plays a significant role in the aspect of image scene classification, building targets recognition and 3D reconstruction of buildings. When there are multiple building targets in one image, the extraction becomes difficult. Since Gestalt principle reflects the features of human visual perception, it has advantages in detecting the man-made targets which are constructed in accordance with certain rules, such as buildings. Therefore, a building area extraction method based on the Gestalt principle is proposed. Firstly, it proposes an algorithm to calculate the salient edge based on texture features and gray feature information, which solves the excessive false extraction problem of the edge extraction method. And using the relevant point to determine the bottom boundary of a building. Secondly, aimed at the contour integrity of the building target, a new graph model method of extracting the closed contour is proposed. Thirdly, mathematical modeling is conducted regarding to the parallelism, symmetry and similarity feature of the Gestalt principle, and the energy function of the closed contour is designed in accordance with Gestalt principle. False closed contours will be deleted by energy function, and this method can precisely extract the closed contour of the building target in the image. Finally, analyzing the relationship between the location of the building area, and combining the regional texture and gray information to extract the building area. Experimental results demonstrate that the method we proposed can extract multiple building areas in the image, and the building area is more accurate and complete in contour than existing building extraction methods.