In inertial/visual integrated navigation system, image registration is the key to achieve image positioning. In the practical application of template matching, there are inevitably changes in rotation and scale between the source image and the template image. In the selection of matching search strategy, the traversal search method is very time-consuming and inefficient, which makes it difficult to meet the requirements of real-time positioning and navigation. To address the problem above, an image registration acceleration algorithm based on improved ORB is proposed in this paper. Image pyramid is introduced in ORB feature detection, which effectively solves the problem of rotation and scale variation between source image and template image. In terms of matching efficiency, a coarse-to-fine hierarchical matching strategy based on forward jumping is adopted, in which a similarity measure threshold is set to quickly determine potential matching regions based on the similarity of the quantified regions to be matched, and then to determine the best matching region. The image matching test results indicate that the proposed image registration algorithm can overcome the effects of scale and rotation variation effectively, and the matching results are robust.