Image registration is an important step in image fusion. In this paper

a new automatic image registration method is presented. First

a small number of feature points are extracted in both images using a Gabor wavelet feature detector. Then

these feature points are matched and the affine transformation between the two images is obtained through a matching technique based on the Hausdorff distance. We choose feature points instead of edges of objects to search for the affine transformation so that the computation load can be decreased largely. On the same time

because the Hausdorff distance is a measure defined between two point sets and does not require to establish an explicit points correspondence between images

it can tolerate errors introduced by the presence of outlier points (noises) as well as the absence of some missing points. Consequently

this registration method can be applied to images with large misalignment. Experiments with synthetic and real images show that this algorithm is efficient.