Image Stitching Based on Anisotropic-Scale Junction

Image stitching often adopts point feature-based matching and global transformation, while point feature only contains location information without local structure information, and global transformation model is only applicable to rotation motion and long-range shooting. When images are taken with viewpoint variations, it may produce obvious registration error, resulting in dreadful image stitching. To this end, this paper proposes a novel image stitching method based on anisotropic-scale junction. Junction integrates the point feature and line feature, which describes the important local geometric structures. Except for point-based registration, junction-based registration also adopts line feature constraint to refine registration. Then combined with local warping, that tolerates local deformations, the proposed approach improves the registration and stitching. At last, some experiments are made to verify the validity of the proposed method.