The goal of traditional image stitching is to achieve seamless stitching. While the goal of thethreedimensional spherical stitching of GSMS remote sensingimagesis to determine the correspondence between each pixeland latitude and longitude. In this paper, a threedimensional spherical stitching of GSMS images is proposed.First of all,initial image registration can be obtained by contour matching based on the topological position relationwith the help of the GSHHG database. Next, the mismatchedpairs are removed byfeature refinement based on the geometric topological consistency. Then, the mapping relations between the images are obtained throughthe fitted polynomial mapping models. After constructing the optimal seamline searching rule,we search an optimal seamline with dynamic programming and realize the threedimensional spherical stitching with the weighted fusionalgorithm.