Image Stitching by Line-guided Local Warping with Global Similarity Constraint

Low-textured image stitching remains a challenging problem. It is difficult to achieve good alignment and is easy to break image structures, due to the insufficient and unreliable point correspondences. Besides, for the viewpoint variations between multiple images, the stitched images suffer from projective distortions. To this end, this paper presents a line-guided local warping with global similarity constraint for image stitching. A two-stage alignment scheme is adopted for good alignment. More precisely, the line correspondence is employed as alignment constraint to guide the accurate estimation of projective warp, then line feature constraints are integrated into mesh-based warping framework to refine the alignment while preserving image structures. To mitigate projectve distortions in non-overlapping regions, we combine global similarity constraint with the projective warps via a weight strategy, so that the final warp slowly changes from projective to similarity across the image. This is also integrated into local multiple homographies model for better parallax handling. Our method is evaluated on a series of images and compared with several other methods. Experiments demonstrate that the proposed method provides convincing stitching performance and outperforms other state-of-the-art methods.