Abstract: An underwater image registration method based on SURF is proposed in this paper to achieve real-time performance and robustness in underwater image mosaic.After a preprocessing on underwater images the SURF algorithm is applied to extract feature points of referring images and registering images.Potential corresponding matching points are allocated with kD tree method.The mapping relationship between images are then obtained using RANSAC and least squares method.Outliers can be removed and at the same time final exact solutions can be obtained based on iterative computation.Finally,after resampling with bilinear interpolation method from registering images,registered images are fused with weighted average to produce underwater image mosaic.Experimental results show that the proposed method obviously improves the image quality.The accuracy of our method is comparable with SIFT-based image registration method while the feature point exaction is about 10 times faster,and registration is about 5 times faster.In addition to high performance achieved,our proposed method is also robust.