Misalignment-eliminated warping image stitching method with grid-based motion statistics matching

Aligning images is one of the main goals of image stitching. Limited by matching accuracy and deformability, the results of current mainstream stitching approaches in large parallax scenes usually contain obvious stitching errors. A misalignment-eliminated warping image stitching based on grid-based motion statistics (GMS) matching is proposed. A matching approach from coarse-to-fine composed of oriented FAST and rotated BRIEF (ORB) and GMS is integrated to provide more accurate and higher number of inliers for subsequent warping. The local homography with global similarity transformation constraint is used to warp the images to achieve initial image alignment. For the projection biases after local warping, a post-processing step based on thin plate spline (TPS) is proposed for further correction. Both qualitative and quantitative comparisons in experiments of challenging cases show that this method can accurately align images while maintaining a natural look at the same time.

Automated summary

This paper proposes a misalignment-eliminated warping image stitching method based on grid-based motion statistics (GMS) matching. The method uses a matching approach from coarse to fine and employs local homography with global similarity constraint for image warping, achieving accurate alignment while maintaining a natural look.