Abstract: In view of the problems of misclassification and omission in building extraction from high-resolution images, the common rectangular contour optimization of single image is not accurate enough due to poor projection, trees and shadow shading, Therefore, an optimization method of multi-temporal building contour is proposed. Firstly, the two high-resolution images are preprocessed by matching, and the initial results of buildings are extracted by shadow classification verification. Then, the minimum area circumscribed rectangles of each building contour in the two images are obtained, and the corresponding corners of the minimum area circumscribed rectangles of the same building are matched. Finally, by comparing the number of building pixels on the corresponding edge of the corresponding minimum area circumscribed rectangle of the two phases of image, the edge with more pixels is selected as the appropriate edge for translation to obtain the optimal edge. After the four edges are the optimal edges, the final contour that conforms to the original shape of the building is obtained by orthogonal lines. Experimental results show that the proposed method improves the accuracy and completeness compared with the traditional results and other contour optimization methods.

Key words: high-resolution remote sensing image, building contour optimization, minimum area enclosing rectangle, straight line orthogonal, optimal edge selection