Building recognition from aerial images combining segmentation and shadow

We propose a novel building detection algorithm for processing high-resolution aerial images. Our algorithm exploits the building-shadow geometric relationship according to lighting models, making it suitable to detect buildings in a more general setting, possibly with irregular shapes. We use image segmentation to provide spatial support for both building and shadow detections. A novel confidence method is developed to label building and shadow segments by jointly reasoning: 1) the likelihood of shadows; 2) building-shadow configuration, and 3) building-building similarity. Our method is tested on a wide range of aerial images. Qualitative and quantitative results demonstrate its effectiveness on detecting and extracting buildings from background clutter.