Registration of images deals with finding geometrical warp parameters between different images with some common intrinsic details. The registration plays one of the most important roles in the super-resolution reconstruction since the latter deals with recovery of one high-resolution image from warped, blurred and decimated replicas of it. Several algorithms were already proposed for the general solution of the image registration. In this paper we concentrate on pure translations case. We use previous basic ideas and results to develop a new registration technique, which uses multiple reference frames, works with hardly aliased images and brings high subpixel precisions.