A novel method is presented for image registration.Firstly, the image data is compressed based on wavelet transform to decrease the time required for subsequent feature point detection, then multi-scale Harris operator is used for detecting feature point set.Fuzzy sets are adopted to reduce the matching range of corresponding feature points.Two sets of feature points with minimum matching error are selected as the control point for getting an affine transformation matrix.And finally image registration is implemented based on the transformation matrix.Experiments show that this method can improve the speed of image registration obviously while decreasing the matching error.