For the requirements of space target detection for image registration, an observation image registration algorithm based on Fourier-Mellin trans-form and Speed Up Robust Feature(SURF) transform was proposed. The transform model, feature detection, feature matching and the matching accuracy related to the algorithm were all discussed. Firstly, the Fourier-Mellin transform was used to calculate the rotation angle, the image to be matched was reversely rotated according to the integer angle of the image. Then, the SURF was used to detect the matching feature points between two images. Finally, the floating point rotation angle and offset were calculated with the least square method. The sum of integer angle and floating point angle is an actual rotation angle of the image to be matched . Furthermore, the relationships between SURF feature point detection and image size, DoH response threshold and scale space decompose layers were analyzed. The experimental results show that the Root Mean Square Error( RMSE) of rotation angle estimation is 0.0077°, and the average value of RMSE of 50 groups of star centroids is 0.1353 pixel. The algorithm can meet the requirements of space target detection.