Registration of Multi-view Images of Planar Surfaces

This paper presents a novel image-based registration method of high-resolution multi-view images of a planar material surface. Contrary to standard registration approaches, this method aligns images based on a true plane of material surface and not on a plane defined by registration marks. It combines the camera calibration and the iterative fitting of desired position and slant of the plane of the surface, image re-registration, and evaluation of alignment of the surface. To optimize images compression performance, we use an error of a compression method as a registration quality evaluation function. The proposed method shows encouraging results on examples of visualization of viewand illuminationdependent textures. In addition to a standard multi-view data registration approach it provides a better multi-view images alignment and thus allows more detailed visualization using the same size of compressed parametrization. Problem Definition Registration of multi-view high-resolution images of planar surfaces: • surface points not lying in a single plane → misaligned after a projective transformation: • registration marks on a separate plane → estimate true surface plane, • feature-based methods unreliable (due to changing illumination & view directions), • registration marks cannot be placed on the surface (easy sample replacement, surface height variations), • higher image resolution → more apparent misalignment, • mechanical compensation → less accurate. Proposed Registration Method Camera calibration + iterative fitting of surface shift and slant → image re-registration → registration quality evaluation: • uniform sampling of search space → global minima → local refinement, • alternating between 3 parameters: plane shift and normal elevation & azimuth, • registration quality criterion: PCA-based compression error. Experimental Results Experiment 1: flat image placed approx. 1mm below the registration plane. • 80 uniformly distributed viewing directions, standard registration proposed registration shift 1.24 mm/elevation 0.29o standard approach proposed approach