An Parallax Image Stitching Method Based on Dynamic Mesh Partition

Abstract

Parallax image stitching is a challenge in the field of image stitching. This paper proposes a method of stitching parallax images based on dynamic mesh partition. Firstly, we extract the matching feature points of the images to be stitched, by which the local zoom scale and the rotation angle between the images can be obtained. At the same time, the images is divided into basic grids, and the matching points can be interpolated to calculate the scaling and rotation angle parameters of each grid. These parameters are then used to calculate the similarity between the meshes and perform a refinement of the mesh based on the similarity. Merge the refined meshes to get the final grid. Finally, we use Moving DLT to calculate the local transformation matrix for each mesh and apply it to complete the image stitching. The experimental results show that our method performs better than the traditional methods.