Building Recognition from High Resolution Image

The recognition of buildings from high resolution remotely-sensed imagery automatically, especially in urban areas, is one of the major difficulties in the remote sensing and the computer vision fields. This paper proposes a new integrated approach, firstly, contourlet and the total variation minimization and oscillating patterns are applied for image preprocessing, secondly, mean shift and automatically seed region growing are used to segment image, and using geometrical characteristics to reject some things that are not buildings, then the phase congruency edge detection is use to get the boundary and the corner of objects, finally, we carry out edge linking and line segment fitting to recognize buildings. Intensive experiments have show that the new method is effective.