Robust image stitching method with line preservation and seamless blending.

Abstract

To address the limitations of traditional homography-based models in accurately aligning overlapping regions during image stitching and the issue of line distortion often caused by mesh warping-based methods, this paper proposes an improved image stitching algorithm. To prevent the bending of linear structures during mesh optimization, straight lines are first extracted using a line segment detection algorithm and then incorporated as constraints in the energy function. This ensures precise alignment of overlapping regions while minimizing line distortion. To further eliminate visible seams in overlapping areas, a graph-cut-based optimal seam blending algorithm is employed to achieve seamless stitching. Experimental results demonstrate that the proposed method effectively aligns overlapping regions, significantly reduces line bending, and produces seamless panoramic images with clear details and minimal global distortion. The algorithm also achieves superior performance in objective evaluation metrics.