Image Mosaicking with Optimized Matching of Global and Local Contents

Photometric inconsistency and geometric misalignment are the main problems of image mosaicking. In our previous studies, we have introduced a method called curvature-domain image stitching, which can successfully reduce the artifacts caused by these two problems. However, it would be computationally heavy if the whole mosaic region uses this method. In this paper, we separate the region into two subregions, while getting their optimized matching with different energy functions. Images in the global subregion are first coarsely matched, followed by the fine matching with the curvature-domain method in the local subregion around the cutting curve. This arrangement can balance the computational effort and the matching quality among the subregions. In the simulation, we will show that both visual quality and computation time are improved under subjective and objective comparisons