High resolution SAR building detection with scene context priming

Building detection is the key step in urban information analysis from high resolution SAR images. Exploring context information is very helpful for object extraction. A method for detecting building units from high resolution SAR images with context information priming is proposed. It tries to use scene category information to prime building detection. Different building detection configurations are used for two different classes of regions. Final result is got by fusing the two different results. The two classes of regions are classified based on features of bright patches. Experiments on a TerraSAR-X intensity image covering part of Wuhan city of China with spatial resolution 1.25m*1.25m have shown the good performance of the method.