We have applied our invariant features to various recognition and registration prob- lems including object recognition, panorama stitching (rotation estimation) and 3D matching (fundamental matrix estimation). Figure 1 shows successful match- ing despite a large change in viewpoint. Figure 5 shows the epipolar geometry computed from invariant feature matches. It can be seen that this epipolar geom- etry is consistent with the images, which are related by camera translation along the optical axis. Figure 6 shows results for object recognition. In this example, we have solved for a homography between the object in the two views. Note the large scale changes between the objects in the images in this case.