This paper presents a LSSIM algorithm for image registration. First extract the feature points from images by classic corner detection method; Then use phase correlation method to estimate the overlapping area between the two images; Then transform the neighborhood area of feature point with Laplace; Last, establish the matching relationship between the two images by the improved SSIM (structural similarity) method. Results show that the LSSIM method can complete a good job of feature point registration. The matching point pairs are not only adequate but also accurate and at the same time have certain robustness to brightness difference, so it can ensure the accuracy of image registration.