Le Zhang, Dong Ren, Zhiyong Huang, Shuanghui Lei, and Chong Zhang


Stitch, parallax, ghosting, decompose, interpolate


Parallax error and ghosting are serious problems in unmanned aerial vehicle (UAV) image stitching. These problems occur even more obviously when stitching together large-parallax images. In this paper, an image stitching method based on projective interpolation is proposed. First, we estimate the homography matrix from the UAV images, obtain its inverse matrix. Second, the two matrices are decomposed and interpolated into sub-matrices. Then these sub-matrices are composed into two new matrices. Finally, the image stitching is finished by warping images using the new matrices. The proposed method disperses the parallax into the two images to help reduce the parallax error. Moreover, the method introduces a seam cutting technique to preserve the geometric structure and further improve local alignment. We conducted three experiments that compared the proposed method with other state-of-the-art methods. The experimental results show that the proposed method can effectively reduce parallax error, eliminate ghosting, and achieve an overall good stitching result.

Important Links:

Go Back