Image stitching using double features-based global similarity constraint and improved seam-cutting

Due to the parallax of input image sequence, it is difficult to achieve accurate alignment, which results in artifacts in image stitching. Besides, since most spatial variation warps extrapolate warps to non-overlapping regions through homography regularization, the stitching results are single-perspective, which causes distortion. To solve these problems, in this paper, we propose a new stitching method. Firstly, feature points and feature lines are employed to improve the accuracy of registration and the naturalness of image warps. And mesh grids are used to guide the warps of the target image. Then, to overcome the distortion problem, we introduce a global similarity constraint. And two weight factors are used to combine the similarity transformation obtained by double features with the projective transformation on the target image, so as to achieve a smooth transition. Finally, after creating a label between pixels in overlapping regions, we design a logical function to optimize the problem of minimizing the label energy function, and search for a stitching seam with a seam-cutting method. Experimental results show that the proposed method can not only effectively suppress the projection distortion in the non-overlapping regions of the target image, but also well remove artifacts in the overlapping regions.