Abstract: This paper analyses the defect of the existing image registration algorithm which is applied in remote sensing image registration field. Addressing the problem about improving the accuracy of different modes for remote sensing image registration, it proposed an assisted multi modal image registration algorithm. Firstly inputting control points on test image and the reference image and using DOG（Difference of Gaussian）to determine the precise coordinates of key points on test images. Secondly, the test image and reference image can get a rough registration by using projection transform and linear least square algorithm. Finally the algorithm automatically adjusts the control points by sub-pixel step according to the registration error and achieves the sub-pixel registration result . The experimental results show that, the algorithm has higher registration accuracy.