This paper proposes a multiresolution image matching framework based on graph cut theory and Hausdorff distance for aerial images with complex scene. After the establishment of Gaussian pyramid image model

we adopt graph cut segmentation in the lowest level and extract robust and integral region boundaries with full consideration of both local and global information. The output region boundaries will be used as candidate curves for further matching. After calculating the statistical information of curves as matching features

we can evaluate the coarse affine transformation parameters using simple correlation measure. The coarse affine transformation parameters estimated can be further used to search in fine levels based on the Hausdorff distance measure. The experiment shows that our method can overcome great distortion and suppress strong noise

and successfully match the images of complex scenes.