Parallax-Tolerant Image Stitching with Optimal Homography

Abstract

Artificial ghosting is a serious problem in image stitching with wide-baseline, which makes it difficult to achieve seamless stitching. A robust parallax-tolerant image stitching method combined with optimal homography matrix selection and seam-cutting processing is proposed. Optimal homography matrix is obtained by analysing feature point distribution and minimizing the registration error. And then, the aligned area is enhanced by the optimal homography matrix, which can facilitate the seam searching. To realize seamless stitching, the luminance difference and the geometrical structure are used in seam-cutting processing. At last, the experimental results show that the artificial ghosting is solved quickly and effectively in the images stitching with large parallax.