A semi-automatic building extraction method based on Mathematical Morphology is proposed in this paper. Firstly, the threshold value is chosen according to the experiment result to get the binary image. And then, the binary image is processed with erosion and dilation operators repetitively to improve the smoothness, connectedness and accuracy of the regions. Moreover, the open and close operators are performed on the image to remove many kinds of noises such as holes, spurs and isolated points. Furthermore, the connected regions are segmented and identified. The building index of each region, which is the ratio of area to perimeter in this paper, is calculated to determine whether each region is a building or not. Finally, the experiment results are given and analyzed to testify the method proposed in this paper.