Based on the scaleinvariant feature transform (SIFT) algorithm, a synthetic aperture radar (SAR) image registration algorithm which can effectively suppress the interference of speckle noise is proposed. Firstly, the anisotropic scale space of the image is constructed by the speckle reducing anisotropic diffusion, so that spackle noise is greatly reduced while prominent structures of the images are preserved, which weakens the impact of speckle noise on feature extraction. Secondly, the bivariate histogram analysis method is improved to refine the initial feature matches obtained by the dualmatching strategy, which eliminates randomly distributed false correspondences. Finally, a reliable method of transformation error analysis between adjacent features is introduced to increase the number of correct matches, which improves the accuracy of the transformation parameters. Experiments demonstrate the applicability of the proposed algorithm to enhance the stability of SIFT features and achieve high registration accuracy. The results show that the proposed algorithm has a high adaptability to speckle noise.