Underwater image stitching using globally optimal local homographies with application to seafloor mosaicing
Bashar Elnashef, Sagi Filin · OCEANS 2021: San Diego – Porto · 2021
We present in this paper a stitching approach for large-scale mosaicing of underwater images. The method we propose adopts a local warp model for the underwater case, which improves the alignment and minimizes local distortions when compared to the more common global wrap approach. This we then improve by adding a global constraint that allows generating optimal warps for all images in a single optimization. This way, the distortions are also minimized globally and linearly in the least-squares sense. A comprehensive evaluation shows that the proposed method generates accurate alignments and provides natural-looking mosaics. Additionally, we propose a sea floor mosaicing pipeline that is based on our stitching method and demonstrate how it yields accurate results even when applied on large datasets.