Abstract: Due to aircraft vibrating, noise and other undesirable factors, image registration can't make the background consistent precisely. To solve this problem, this paper presents a method with feature point accuracy purification for image registration. In the corner extraction part, it lowers corner redundancy by controlling the amount and accuracy of the feature point; and in the corner matching part, it remove false matching points by removing the points with larger matching error. Furthermore, establish the Gaussian pyramid about resolution for the sake of reducing algorithm complexity. Experimental results show that the improved algorithm of image registration based on the feature point accuracy purification has better results than that without purification of the feature point, and the calculated amount is reduced to about 28% of the original.