3D building similarity for a random single-view-image pair based on a local 3D shape

Abstract

In this paper, 3D building shape similarities using graph similarity calculations are discussed. Moreover, a method for characterizing and measuring building shapes in urban landscape analysis and visual perceptual orientation is developed by constructing a building shape graph with geometric invariance for shape graph measurements. Specifically, we proposed a local shape feature at intersection points with 3D invariance. Then, we suggested an approach to obtain the global shape graph similarity between two buildings using graph kernels and thus obtained the shape similarity between two buildings. The performance is evaluated with different sets of experiments. The results show that the similarity of our local shape features at the same point is greater than 0.70, the global shape graph structure can clearly identify the differences between the same buildings due to occlusion and avoid the interference of textures between different buildings, and the similarity measure approach achieves accuracy is more than 89% on a large number of triplet validation data, which reflects excellent applicability.

Keywords

• 3D building similarity
• Local shape feature
• Global shape graph
• Graph kernels
• Graph similarity measure