Image superresolution reconstruction is a process of reconstructing a highresolution image from a set of lowresolution images of the same scene. It can be achieved in two steps: registering all low-resolution images at sub-pixel accuracy, and using the registered low-resolution images to reconstruct a superresolution image. This paper proposes a novel method of subpixel image registration. Preliminary registration is done using the Vandewalle’s method with only the approximation of subimages in the Haar domain. Fine registration is achieved using an improved Keren’s method. Registration is then conducted in a pyramid fashion, and the registered results of all layers are accumulated. In terms of registration accuracy and reconstruction quality, experiments demonstrate that the proposed method has better performance as compared to the improved Keren’s method and Vanderwalle’s method even though the shift and rotation are large.