Structure propagation for deformable image stitching

Stitching of images is often challenging due to lack or insufficient amount of information shared by sub-images to be stitched. Imaging artifacts such as distortions complicate the process further by decreasing the reliability of available information in overlap region. Without taking these into account, a stitching approach might violate global continuity and smoothness of structures across overlap region. In this paper, we propose a new registration approach for deformable stitching of sub-images in such scenarios. Using tensor voting, we propagate structures throughout the overlap region, creating a structural map to guide the optimization towards an optimum that will ensure the global structural continuity after stitching. Experimental results on synthetic and real images in 2D and 3D demonstrate robust performance of the proposed registration technique.