A Graph Matching and Energy Minimization Based Algorithm for Lunar Surface Image Mosaic

This paper concerns the problem of lunar surface image mosaic, including both image registration and image stitching procedures. A wide viewing composite obtained by mosaic technique plays an important role in many lunar rover's operations. Considering particular characters in lunar surface images, such as large geometrical deformations, significant illumination differences and repeated patterns, previous image mosaic techniques often fail to create a qualified composite. In this paper, a novel algorithm is introduced to tackle the lunar surface image mosaic problem. Specifically, in the image registration procedure, to deal with the misregistration problem caused by large geometrical deformation and repeated patterns, structural information is introduced to solve the feature correspondence by formulating it as a graph matching problem. In the image stitching procedure, an energy minimization method is proposed based on the graduated nonconvexity and concavity procedure (GNCCP), to handle the visible seams caused by illumination differences and ghosting problem caused by large parallax in the overlapped area. Comparative experiments on real lunar surface images acquired by Yutu rover and Apollo image gallery validate the effectiveness of the proposed method.