Feature Layering and Multi-Plane Warping for Image Stitching

Enhancing the alignment of overlapping regions after image warping in image stitching is an important issue. In this paper, we propose a stitching method based on feature layering and multi-plane warping. First, the detected feature points are subjected to layered selection. However, some outliers may still exist among the retained inliers. Therefore, further evaluation and filtering are applied to these outlier noise points. This filtering process preserves more accurate feature inliers. Next, we detect linear structures as features and compute multilayer homographies based on multi-layer feature inliers, as well as homography based on feature lines. We then utilize a weighted summation to calculate the global homography based on point-line dual features to optimize the warping transformation. Experiments show that the proposed method performs well in enhancing alignment and reducing ghosting and misalignment based on existing publicly available datasets.